From 61bca0d6425114d57ec2e69b2f3762811e52c4e5 Mon Sep 17 00:00:00 2001
From: everyoneexe <mansurozdemir41@gmail.com>
Date: Tue, 4 Aug 2026 03:06:05 +0200
Subject: [PATCH 1/3] docs: add design spec for custom model settings override

Custom model IDs on router providers (e.g. anthropic/claude-sonnet-4-6 on
OpenRouter) render a context window of 1 and a 7000% usage figure when the
model list is unavailable. Spec covers three defects: the TaskHeader `|| 1`
fallback with no upper clamp, webview/host divergence in model resolution,
and the absence of any override UI outside the OpenAI-compatible provider.

Design: one `customModelInfo` overlay field on the base provider schema, one
shared `applyCustomModelInfo` helper bound at both resolution layers, display
hardening independent of any override, and a collapsible settings panel for
the five router providers.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---
 ...2026-08-04-custom-model-settings-design.md | 301 ++++++++++++++++++
 1 file changed, 301 insertions(+)
 create mode 100644 docs/superpowers/specs/2026-08-04-custom-model-settings-design.md

diff --git a/docs/superpowers/specs/2026-08-04-custom-model-settings-design.md b/docs/superpowers/specs/2026-08-04-custom-model-settings-design.md
new file mode 100644
index 0000000000..fe9bc538ee
--- /dev/null
+++ b/docs/superpowers/specs/2026-08-04-custom-model-settings-design.md
@@ -0,0 +1,301 @@
+# Custom Model Settings Override
+
+**Date:** 2026-08-04
+**Status:** Approved, ready for implementation plan
+
+## Problem
+
+A user selects a model ID that is not present in a router provider's fetched
+model list — for example typing `anthropic/claude-sonnet-4-6` into the
+OpenRouter model picker via the "use custom model" affordance
+(`webview-ui/src/components/settings/ModelPicker.tsx:277`). Three distinct
+defects follow.
+
+### Defect 1 — context window collapses to `1`, percentage renders as 7000%
+
+`webview-ui/src/components/chat/TaskHeader.tsx:72`:
+
+```ts
+const contextWindow = model?.contextWindow || 1
+```
+
+When `useSelectedModel` cannot resolve the model, `model` is `undefined` and
+`contextWindow` becomes `1`. That value flows into the percentage at
+`TaskHeader.tsx:253-258`:
+
+```ts
+const availableInputSpace = contextWindow - reservedForOutput
+const percentage =
+    availableInputSpace > 0
+        ? Math.round(((contextTokens || 0) / availableInputSpace) * 100)
+        : 0
+```
+
+With `contextWindow === 1` and `reservedForOutput === 0`, `availableInputSpace`
+is `1`, so the percentage equals `contextTokens * 100`. 70 context tokens
+render as **7000%**. There is no upper clamp on this path.
+
+**Precise trigger.** For OpenRouter an unknown ID is normally rewritten to the
+default model by `getValidatedModelId`
+(`webview-ui/src/components/ui/hooks/useSelectedModel.ts:56`), which yields a
+valid `info`. The `undefined` case therefore arises when the router model list
+is empty rather than merely missing the ID: no API key configured, a failed or
+in-flight fetch, or offline. In that state the default-model lookup also misses,
+`info` is `undefined`, and the `|| 1` fallback produces both the "token limit
+shows 1" symptom and the 7000% reading. They are two faces of one fault.
+
+### Defect 2 — webview and extension host disagree on the model
+
+`getValidatedModelId` silently substitutes the provider default when the
+configured ID is absent from the list, while `openRouterModelId` continues to
+hold the user's typed value. The extension host does not perform the same
+substitution — `src/api/providers/openrouter.ts:554`:
+
+```ts
+let info = this.models[id] ?? openRouterDefaultModelInfo
+```
+
+The host sends the user's real ID with a 200K-context default profile; the
+webview displays a different model entirely. Requests may succeed while the UI
+describes something else.
+
+### Defect 3 — no override UI outside the OpenAI-compatible provider
+
+`openAiCustomModelInfo` (`packages/types/src/provider-settings.ts:245`) is the
+only user-facing way to supply `contextWindow` / `maxTokens`, and it is wired
+solely to the `openai` provider's settings panel
+(`webview-ui/src/components/settings/providers/OpenAICompatible.tsx:286-347`).
+OpenRouter, Requesty, Unbound, Vercel AI Gateway and Zoo Gateway offer no
+equivalent, so a custom model on those providers can never be given correct
+token limits.
+
+## Goals
+
+1. Let the user override context window and max output tokens for any model on
+   the router providers, and have that override govern both the UI and the real
+   request/truncation path.
+2. Ensure the UI never displays a nonsensical figure when no override is set.
+
+## Non-goals
+
+- Editing per-token pricing. Overridden prices would corrupt cost reporting;
+  that is separate work.
+- Migrating or removing `openAiCustomModelInfo`. It keeps working unchanged.
+- Reworking `ModelPicker`'s custom-model entry flow.
+
+## Architecture
+
+Two layers resolve model info independently and must not diverge:
+
+| Layer | Resolver |
+|---|---|
+| Webview | `getSelectedModel()` in `useSelectedModel.ts:132` |
+| Extension host | each provider's `getModel()` (30 implementations) |
+
+An override applied to only one layer would fix the display while leaving
+context truncation wrong. The design therefore applies one shared helper at both
+layers, each through a single chokepoint.
+
+Two facts from the codebase make the host-side chokepoint viable: there is
+exactly one factory, `buildApiHandler` (`src/api/index.ts:153`), and no
+`instanceof <X>Handler` check exists anywhere in `src/`. A wrapper around the
+returned handler is therefore safe.
+
+### Data model
+
+Add one field to `baseProviderSettingsSchema`
+(`packages/types/src/provider-settings.ts:176`):
+
+```ts
+customModelInfo: modelInfoSchema.partial().nullish(),
+```
+
+`partial()` is deliberate. The field is an **overlay**, not a replacement: a user
+who sets only `contextWindow` keeps the fetched values for price, image support
+and reasoning. Placing it on the base schema means every provider inherits it,
+avoiding the five near-identical fields that a per-provider approach would need.
+
+`openAiCustomModelInfo` remains as-is. Where both are present, `customModelInfo`
+is applied second and wins on the fields it defines.
+
+### Shared helper
+
+In `packages/types` (importable by both webview and host):
+
+```ts
+applyCustomModelInfo(
+    info: ModelInfo | undefined,
+    settings: { customModelInfo?: Partial<ModelInfo> | null } | undefined,
+): ModelInfo | undefined
+```
+
+Behaviour:
+
+- `info` present → return `info` with the override's **defined and valid** keys
+  merged over it.
+- `info` absent but the override supplies a positive `contextWindow` → synthesise
+  a `ModelInfo` from a synthesis base plus the override. This is what makes a
+  genuinely unknown model usable.
+- Neither → return `undefined`, preserving today's "invalid selection" signal.
+
+The synthesis base is defined locally rather than reusing
+`openAiModelInfoSaneDefaults`, whose `maxTokens: -1` sentinel
+(`packages/types/src/providers/openai.ts:692-693`) would propagate a negative
+value into arithmetic:
+
+```ts
+const CUSTOM_MODEL_SYNTHESIS_BASE = {
+    maxTokens: undefined,
+    supportsImages: false,
+    supportsPromptCache: false,
+} satisfies Partial<ModelInfo>
+```
+
+`contextWindow` is deliberately absent from the base: synthesis only runs when
+the override supplies a positive one, so the merged result always has a real
+value and never a fabricated default.
+
+Leaving `maxTokens` undefined is safe rather than lossy. `getModelMaxOutputTokens`
+(`src/shared/api.ts:131-133`) supplies `ANTHROPIC_DEFAULT_MAX_TOKENS` whenever the
+model ID contains `claude` and `maxTokens` is absent — which covers the reported
+`anthropic/claude-sonnet-4-6` case. For non-Anthropic IDs it returns `undefined`
+(line 158-160), which `TaskHeader` already handles by reserving nothing.
+
+A key is treated as "valid" when it is not `undefined`/`null`, and — for the
+numeric fields `contextWindow` and `maxTokens` — is a finite number greater than
+zero. Invalid entries are dropped, never coerced to `0`, because
+`contextWindow: 0` would reproduce the original division fault.
+
+### Integration points
+
+**Webview** — apply the helper to the `{ id, info }` produced by the ternary at
+`useSelectedModel.ts:98-113`, not to `getSelectedModel()`'s return. That ternary
+has three branches: the resolved call, a `kimi-code` fallback, and a
+not-ready/invalid-provider fallback that yields `info: undefined`. The override
+must cover all three — the third is precisely the still-loading state that
+produces the reported symptom, and `getSelectedModel()` is not called there at
+all. Applying it after the ternary covers every branch and leaves the 30 `switch`
+cases untouched.
+
+**Host** — in `buildApiHandler`, wrap the constructed handler in a `Proxy` that
+decorates `getModel()` and forwards everything else. Forwarding uses
+`Reflect.get(target, prop, target)` — passing `target` rather than the proxy as
+receiver, so private class fields continue to resolve. All twelve
+`this.api.getModel().info` consumers in `Task.ts` inherit the corrected value,
+including the context-window-exceeded and condense paths.
+
+### Display hardening
+
+Independent of any override, so the UI is correct when the user sets nothing:
+
+- `TaskHeader.tsx:72` — drop `|| 1`. When no context window is known, skip
+  rendering the percentage entirely rather than printing a fabricated number.
+- `TaskHeader.tsx:253-258` — clamp the upper bound with `Math.min(100, …)` and
+  render at/over 100% in a warning colour. Keep the existing
+  `availableInputSpace > 0` guard: it is the lower bound, and an over-large
+  `maxTokens` override can still drive `availableInputSpace` to zero or below.
+- `useSelectedModel.ts:51-57` — stop substituting the provider default for a
+  configured-but-unlisted ID on the router providers. The condition is *the
+  configured ID is absent from the list*, which covers both an empty list and a
+  populated list that lacks the user's custom ID; the current guard conflates
+  them. The litellm case (lines 178-189) is the in-repo precedent for returning
+  the configured ID untouched.
+
+  This aligns the webview with the host, which never substitutes — closing
+  Defect 2's divergence. It does not make the two produce identical `info`: the
+  host still falls back to `openRouterDefaultModelInfo` (200K) at
+  `openrouter.ts:554` while the webview yields `undefined`. Full convergence is
+  what the shared helper delivers once an override exists, and is why the helper
+  must be bound at both layers rather than the webview alone.
+
+  Callers that assume a non-empty, listed ID must be checked. `ModelPicker`
+  already tolerates it: `modelIds` explicitly retains `selectedModelId`
+  (lines 122-127) and the initialization effect at 187-194 only fires when
+  `selectedModelId` is falsy, so a preserved custom ID is displayed rather than
+  overwritten.
+
+### UI
+
+New shared component `CustomModelInfoSettings.tsx`, following the field pattern
+already established in `OpenAICompatible.tsx` (text field, green/red border
+validation, label plus description). Rendered beneath `ModelPicker` for the
+router providers: OpenRouter, Requesty, Unbound, Vercel AI Gateway, Zoo Gateway.
+
+Collapsible, collapsed by default. It auto-expands, with an explanatory note,
+when the selected model has no resolved info — the exact situation this feature
+addresses.
+
+Fields: **context window**, **max output tokens**, **supportsImages**,
+**supportsPromptCache**. A "reset to detected values" control clears the
+override.
+
+New i18n keys under `settings:providers.customModelInfo.*` in
+`webview-ui/src/i18n/locales/en/settings.json`. Only English is authored; other
+locales fall back until translated.
+
+## Error handling
+
+| Input | Result |
+|---|---|
+| Empty string | Key omitted from overlay |
+| `NaN` / non-numeric | Key omitted, red border |
+| `<= 0` | Key omitted, red border |
+| Valid positive integer | Applied, green border |
+
+`maxTokens` exceeding `contextWindow` is accepted but flagged with an inline
+warning. The 20% context-window clamp in `getModelMaxOutputTokens`
+(`src/shared/api.ts:154`) is **not** a reliable backstop here — three earlier
+branches return before reaching it: reasoning-budget models (line 117), Anthropic
+contexts with `supportsReasoningBudget` or absent `maxTokens` (lines 126-133),
+and `supportsMaxTokens` models honouring an explicit `modelMaxTokens` (line 138).
+The first two are exactly the `anthropic/claude-*` path in this bug report.
+
+Since the clamp cannot be relied on, the inline warning is the actual guard, and
+`TaskHeader`'s `availableInputSpace` must tolerate `reservedForOutput >=
+contextWindow`. Its existing `> 0` guard already returns `0%` rather than a
+negative percentage; the display-hardening change must preserve that guard rather
+than replace it with the new `Math.min(100, …)` clamp.
+
+## Testing
+
+- `applyCustomModelInfo` unit tests: overlay onto existing info; synthesis from
+  absent info; empty/invalid/zero/negative input dropped rather than coerced;
+  `undefined` returned when nothing is available.
+- `TaskHeader` regression test: with `info === undefined`, assert no `7000%`-class
+  output — the percentage element is absent. This is the lock on the reported bug.
+- `TaskHeader` clamp test: `contextTokens` exceeding the window renders `100%`,
+  not more.
+- `useSelectedModel` tests: an empty router model list preserves the configured
+  custom ID rather than substituting the default; a *populated* list that lacks
+  the configured ID also preserves it. The second case is the one the current
+  guard gets wrong.
+- `useSelectedModel` test: the override applies in the not-ready branch (router
+  models still loading), where `getSelectedModel()` is never called.
+- `buildApiHandler` proxy test: `getModel()` reflects the override while other
+  methods and private field access remain intact.
+- `CustomModelInfoSettings` component tests: validation borders, persistence,
+  reset, auto-expansion when info is unresolved.
+
+## Files affected
+
+| File | Change |
+|---|---|
+| `packages/types/src/provider-settings.ts` | Add `customModelInfo` to base schema |
+| `packages/types/src/model.ts` (or sibling) | Add `applyCustomModelInfo` + synthesis base |
+| `webview-ui/src/components/ui/hooks/useSelectedModel.ts` | Apply helper after the ternary (98-113); stop substituting the default for an unlisted ID |
+| `src/api/index.ts` | Proxy-wrap handler in `buildApiHandler` |
+| `webview-ui/src/components/chat/TaskHeader.tsx` | Remove `\|\| 1`; clamp; conditional render |
+| `webview-ui/src/components/settings/CustomModelInfoSettings.tsx` | New component |
+| `webview-ui/src/components/settings/providers/{OpenRouter,Requesty,Unbound,VercelAiGateway,ZooGateway}.tsx` | Mount component |
+| `webview-ui/src/i18n/locales/en/settings.json` | New keys |
+
+## Risks
+
+- **Proxy overhead** — `getModel()` is called frequently (twelve sites in
+  `Task.ts` alone). The decoration is a shallow object spread over a plain
+  object; negligible, but the overlay should not be recomputed per call beyond
+  that.
+- **Stale override after model switch** — an override set for one custom model
+  persists when the user picks a different one. Accepted: the reset control and
+  the collapsed-by-default panel keep this visible. Auto-clearing on model change
+  risks discarding deliberate configuration.

From 3f7641f4162c8ff15bec110da9d20799032ce3df Mon Sep 17 00:00:00 2001
From: everyoneexe <mansurozdemir41@gmail.com>
Date: Tue, 4 Aug 2026 17:28:39 +0200
Subject: [PATCH 2/3] feat: support custom model info overrides

---
 .../src/__tests__/custom-model-info.test.ts   |  68 ++++++
 packages/types/src/model.ts                   |  74 ++++++
 packages/types/src/provider-settings.ts       |   2 +
 .../providers/__tests__/openrouter.spec.ts    |  20 ++
 src/api/providers/__tests__/requesty.spec.ts  |  18 ++
 src/api/providers/__tests__/unbound.spec.ts   |  21 ++
 src/api/providers/openrouter.ts               |   7 +-
 src/api/providers/requesty.ts                 |  12 +-
 src/api/providers/router-provider.ts          |  16 +-
 src/api/providers/unbound.ts                  |  12 +-
 webview-ui/src/components/chat/TaskHeader.tsx | 142 +++++++-----
 .../chat/__tests__/TaskHeader.spec.tsx        |  22 ++
 .../src/components/settings/ApiOptions.tsx    |   5 +
 .../settings/CustomModelInfoSettings.tsx      | 218 ++++++++++++++++++
 .../CustomModelInfoSettings.spec.tsx          |  67 ++++++
 .../settings/providers/OpenRouter.tsx         |   9 +
 .../settings/providers/Requesty.tsx           |   9 +
 .../components/settings/providers/Unbound.tsx |   9 +
 .../settings/providers/VercelAiGateway.tsx    |   9 +
 .../settings/providers/ZooGateway.tsx         |  11 +-
 .../providers/__tests__/ZooGateway.spec.tsx   |   7 +-
 .../hooks/__tests__/useSelectedModel.spec.ts  |  71 +++---
 .../components/ui/hooks/useSelectedModel.ts   |  73 +++++-
 webview-ui/src/i18n/locales/en/settings.json  |  23 ++
 24 files changed, 811 insertions(+), 114 deletions(-)
 create mode 100644 packages/types/src/__tests__/custom-model-info.test.ts
 create mode 100644 webview-ui/src/components/settings/CustomModelInfoSettings.tsx
 create mode 100644 webview-ui/src/components/settings/__tests__/CustomModelInfoSettings.spec.tsx

diff --git a/packages/types/src/__tests__/custom-model-info.test.ts b/packages/types/src/__tests__/custom-model-info.test.ts
new file mode 100644
index 0000000000..1d2b3da830
--- /dev/null
+++ b/packages/types/src/__tests__/custom-model-info.test.ts
@@ -0,0 +1,68 @@
+import { applyCustomModelInfo, customModelInfoSchema, type ModelInfo } from "../model.js"
+import { providerIdentifiers, providerSettingsSchemaDiscriminated } from "../index.js"
+
+describe("custom model info", () => {
+	it("overlays only supported metadata and preserves provider-owned fields", () => {
+		const model: ModelInfo = {
+			maxTokens: 4096,
+			contextWindow: 8192,
+			supportsImages: false,
+			supportsPromptCache: false,
+			inputPrice: 0.1,
+			outputPrice: 0.2,
+			description: "Provider metadata",
+		}
+
+		expect(
+			applyCustomModelInfo(model, {
+				customModelInfo: {
+					contextWindow: 128_000,
+					maxTokens: 16_384,
+					supportsImages: true,
+					supportsPromptCache: true,
+				},
+			}),
+		).toEqual({
+			...model,
+			contextWindow: 128_000,
+			maxTokens: 16_384,
+			supportsImages: true,
+			supportsPromptCache: true,
+		})
+	})
+
+	it("does not synthesize model info without a valid context window", () => {
+		expect(
+			applyCustomModelInfo(undefined, {
+				customModelInfo: {
+					contextWindow: 0,
+					maxTokens: -1,
+					supportsImages: true,
+				},
+			}),
+		).toBeUndefined()
+	})
+
+	it("synthesizes safe defaults when only a context window is supplied", () => {
+		expect(
+			applyCustomModelInfo(undefined, {
+				customModelInfo: { contextWindow: 64_000, supportsImages: true },
+			}),
+		).toEqual({
+			maxTokens: undefined,
+			contextWindow: 64_000,
+			supportsImages: true,
+			supportsPromptCache: false,
+		})
+	})
+
+	it("rejects unsupported pricing fields in the persisted override schema", () => {
+		expect(customModelInfoSchema.safeParse({ contextWindow: 64_000, inputPrice: 1 }).success).toBe(false)
+		expect(
+			providerSettingsSchemaDiscriminated.safeParse({
+				apiProvider: providerIdentifiers.openrouter,
+				customModelInfo: { contextWindow: 64_000, outputPrice: 1 },
+			}).success,
+		).toBe(false)
+	})
+})
diff --git a/packages/types/src/model.ts b/packages/types/src/model.ts
index 9fbf9e358b..b7c74219c0 100644
--- a/packages/types/src/model.ts
+++ b/packages/types/src/model.ts
@@ -183,6 +183,80 @@ export const modelInfoSchema = z.object({
 
 export type ModelInfo = z.infer<typeof modelInfoSchema>
 
+/**
+ * User-supplied metadata for a model whose discovered metadata is incomplete
+ * or unavailable. This is intentionally narrower than ModelInfo: prices and
+ * other accounting fields must remain provider-owned.
+ */
+export const customModelInfoSchema = z
+	.object({
+		maxTokens: z.number().int().positive().optional(),
+		contextWindow: z.number().int().positive().optional(),
+		supportsImages: z.boolean().optional(),
+		supportsPromptCache: z.boolean().optional(),
+	})
+	.strict()
+
+export type CustomModelInfo = z.infer<typeof customModelInfoSchema>
+
+export type CustomModelInfoSettings = {
+	customModelInfo?: Partial<CustomModelInfo> | null
+}
+
+const isPositiveInteger = (value: unknown): value is number =>
+	typeof value === "number" && Number.isSafeInteger(value) && value > 0
+
+/**
+ * Applies the user metadata overlay without allowing invalid values to enter
+ * model arithmetic or cost/capability fields outside the supported override.
+ * When no discovered info exists, a context-window override is required to
+ * synthesize a usable ModelInfo.
+ */
+export const applyCustomModelInfo = (
+	info: ModelInfo | undefined,
+	settings: CustomModelInfoSettings | undefined,
+): ModelInfo | undefined => {
+	const override = settings?.customModelInfo
+
+	if (!override) {
+		return info
+	}
+
+	const validOverride: CustomModelInfo = {}
+
+	if (isPositiveInteger(override.contextWindow)) {
+		validOverride.contextWindow = override.contextWindow
+	}
+
+	if (isPositiveInteger(override.maxTokens)) {
+		validOverride.maxTokens = override.maxTokens
+	}
+
+	if (typeof override.supportsImages === "boolean") {
+		validOverride.supportsImages = override.supportsImages
+	}
+
+	if (typeof override.supportsPromptCache === "boolean") {
+		validOverride.supportsPromptCache = override.supportsPromptCache
+	}
+
+	if (info) {
+		return Object.keys(validOverride).length > 0 ? { ...info, ...validOverride } : info
+	}
+
+	if (!validOverride.contextWindow) {
+		return undefined
+	}
+
+	return {
+		maxTokens: undefined,
+		contextWindow: validOverride.contextWindow,
+		supportsImages: false,
+		supportsPromptCache: false,
+		...validOverride,
+	}
+}
+
 export type ModelRecord = Record<string, ModelInfo>
 
 export type RouterModels = Record<DynamicProvider | LocalProvider, ModelRecord>
diff --git a/packages/types/src/provider-settings.ts b/packages/types/src/provider-settings.ts
index e17cd5ddbc..344ed93daa 100644
--- a/packages/types/src/provider-settings.ts
+++ b/packages/types/src/provider-settings.ts
@@ -2,6 +2,7 @@ import { z } from "zod"
 
 import {
 	modelInfoSchema,
+	customModelInfoSchema,
 	openAiCodexServiceTierSchema,
 	reasoningEffortSettingSchema,
 	verbosityLevelsSchema,
@@ -185,6 +186,7 @@ const baseProviderSettingsSchema = z.object({
 	reasoningEffort: reasoningEffortSettingSchema.optional(),
 	modelMaxTokens: z.number().optional(),
 	modelMaxThinkingTokens: z.number().optional(),
+	customModelInfo: customModelInfoSchema.nullish(),
 
 	// Model verbosity.
 	verbosity: verbosityLevelsSchema.optional(),
diff --git a/src/api/providers/__tests__/openrouter.spec.ts b/src/api/providers/__tests__/openrouter.spec.ts
index 254cd1dad4..6f5d42ab10 100644
--- a/src/api/providers/__tests__/openrouter.spec.ts
+++ b/src/api/providers/__tests__/openrouter.spec.ts
@@ -139,6 +139,26 @@ describe("OpenRouterHandler", () => {
 			})
 		})
 
+		it("applies custom metadata before deriving request parameters", async () => {
+			const handler = new OpenRouterHandler({
+				...mockOptions,
+				customModelInfo: {
+					contextWindow: 100_000,
+					maxTokens: 10_000,
+					supportsImages: false,
+					supportsPromptCache: false,
+				},
+			})
+
+			const result = await handler.fetchModel()
+
+			expect(result.info.contextWindow).toBe(100_000)
+			expect(result.info.maxTokens).toBe(10_000)
+			expect(result.info.supportsImages).toBe(false)
+			expect(result.info.supportsPromptCache).toBe(false)
+			expect(result.maxTokens).toBe(10_000)
+		})
+
 		it("returns default model info when options are not provided", async () => {
 			const handler = new OpenRouterHandler({})
 			const result = await handler.fetchModel()
diff --git a/src/api/providers/__tests__/requesty.spec.ts b/src/api/providers/__tests__/requesty.spec.ts
index 77adb8724f..8d0d203d1d 100644
--- a/src/api/providers/__tests__/requesty.spec.ts
+++ b/src/api/providers/__tests__/requesty.spec.ts
@@ -158,6 +158,24 @@ describe("RequestyHandler", () => {
 			})
 		})
 
+		it("applies custom metadata before deriving request parameters", async () => {
+			const handler = new RequestyHandler({
+				...mockOptions,
+				customModelInfo: {
+					contextWindow: 100_000,
+					maxTokens: 10_000,
+					supportsImages: false,
+					supportsPromptCache: false,
+				},
+			})
+
+			const result = await handler.fetchModel()
+
+			expect(result.info.contextWindow).toBe(100_000)
+			expect(result.info.maxTokens).toBe(10_000)
+			expect(result.maxTokens).toBe(10_000)
+		})
+
 		it("returns default model info when options are not provided", async () => {
 			const handler = new RequestyHandler({})
 			const result = await handler.fetchModel()
diff --git a/src/api/providers/__tests__/unbound.spec.ts b/src/api/providers/__tests__/unbound.spec.ts
index 0e18c4b175..b9d7f8acb2 100644
--- a/src/api/providers/__tests__/unbound.spec.ts
+++ b/src/api/providers/__tests__/unbound.spec.ts
@@ -38,6 +38,27 @@ describe("UnboundHandler", () => {
 		vi.clearAllMocks()
 	})
 
+	it("applies custom metadata before deriving request parameters", async () => {
+		const handler = new UnboundHandler({
+			unboundApiKey: "test-key",
+			unboundModelId: "openai/gpt-4o",
+			customModelInfo: {
+				contextWindow: 100_000,
+				maxTokens: 10_000,
+				supportsImages: false,
+				supportsPromptCache: true,
+			},
+		})
+
+		const result = await handler.fetchModel()
+
+		expect(result.info.contextWindow).toBe(100_000)
+		expect(result.info.maxTokens).toBe(10_000)
+		expect(result.info.supportsImages).toBe(false)
+		expect(result.info.supportsPromptCache).toBe(true)
+		expect(result.maxTokens).toBe(10_000)
+	})
+
 	it("identifies itself as Zoo Code in the Unbound request headers", () => {
 		new UnboundHandler({
 			unboundApiKey: "test-key",
diff --git a/src/api/providers/openrouter.ts b/src/api/providers/openrouter.ts
index 3e59b4360b..4e7a854bfb 100644
--- a/src/api/providers/openrouter.ts
+++ b/src/api/providers/openrouter.ts
@@ -10,6 +10,7 @@ import {
 	OPENROUTER_DEFAULT_PROVIDER_NAME,
 	OPEN_ROUTER_PROMPT_CACHING_MODELS,
 	DEEP_SEEK_DEFAULT_TEMPERATURE,
+	applyCustomModelInfo,
 } from "@roo-code/types"
 import { TelemetryService } from "@roo-code/telemetry"
 
@@ -551,15 +552,19 @@ export class OpenRouterHandler extends BaseProvider implements SingleCompletionH
 
 	override getModel() {
 		const id = this.options.openRouterModelId ?? openRouterDefaultModelId
-		let info = this.models[id] ?? openRouterDefaultModelInfo
+		const discoveredInfo = this.models[id]
+		let hasDiscoveredInfo = discoveredInfo !== undefined
+		let info = discoveredInfo ?? openRouterDefaultModelInfo
 
 		// If a specific provider is requested, use the endpoint for that provider.
 		if (this.options.openRouterSpecificProvider && this.endpoints[this.options.openRouterSpecificProvider]) {
 			info = this.endpoints[this.options.openRouterSpecificProvider]
+			hasDiscoveredInfo = true
 		}
 
 		// Apply tool preferences for models accessed through routers (OpenAI, Gemini)
 		info = applyRouterToolPreferences(id, info)
+		info = applyCustomModelInfo(hasDiscoveredInfo ? info : undefined, this.options) ?? info
 
 		const isDeepSeekR1 = id.startsWith("deepseek/deepseek-r1") || id === "perplexity/sonar-reasoning"
 
diff --git a/src/api/providers/requesty.ts b/src/api/providers/requesty.ts
index 5753660de5..19531e47db 100644
--- a/src/api/providers/requesty.ts
+++ b/src/api/providers/requesty.ts
@@ -1,7 +1,13 @@
 import { Anthropic } from "@anthropic-ai/sdk"
 import OpenAI from "openai"
 
-import { type ModelInfo, type ModelRecord, requestyDefaultModelId, requestyDefaultModelInfo } from "@roo-code/types"
+import {
+	applyCustomModelInfo,
+	type ModelInfo,
+	type ModelRecord,
+	requestyDefaultModelId,
+	requestyDefaultModelInfo,
+} from "@roo-code/types"
 
 import type { ApiHandlerOptions } from "../../shared/api"
 import { calculateApiCostOpenAI } from "../../shared/cost"
@@ -80,11 +86,13 @@ export class RequestyHandler extends BaseProvider implements SingleCompletionHan
 
 	override getModel() {
 		const id = this.options.requestyModelId ?? requestyDefaultModelId
-		const cachedInfo = this.models[id] ?? requestyDefaultModelInfo
+		const discoveredInfo = this.models[id]
+		const cachedInfo = discoveredInfo ?? requestyDefaultModelInfo
 		let info: ModelInfo = cachedInfo
 
 		// Apply tool preferences for models accessed through routers (OpenAI, Gemini)
 		info = applyRouterToolPreferences(id, info)
+		info = applyCustomModelInfo(discoveredInfo ? info : undefined, this.options) ?? info
 
 		const params = getModelParams({
 			format: "anthropic",
diff --git a/src/api/providers/router-provider.ts b/src/api/providers/router-provider.ts
index cbdd49e58b..6b0eb8b4e4 100644
--- a/src/api/providers/router-provider.ts
+++ b/src/api/providers/router-provider.ts
@@ -1,6 +1,6 @@
 import OpenAI from "openai"
 
-import { type ModelInfo, type ModelRecord } from "@roo-code/types"
+import { applyCustomModelInfo, type ModelInfo, type ModelRecord } from "@roo-code/types"
 
 import { ApiHandlerOptions, RouterName } from "../../shared/api"
 
@@ -58,6 +58,14 @@ export abstract class RouterProvider extends BaseProvider {
 
 	private modelFetchPromise?: Promise<{ id: string; info: ModelInfo }>
 
+	private resolveModelInfo(info: ModelInfo | undefined, fallback: ModelInfo): ModelInfo {
+		if (this.name !== "vercel-ai-gateway" && this.name !== "zoo-gateway") {
+			return info ?? fallback
+		}
+
+		return applyCustomModelInfo(info, this.options) ?? fallback
+	}
+
 	public async fetchModel() {
 		if (Object.keys(this.models).length > 0) {
 			return this.getModel()
@@ -96,7 +104,7 @@ export abstract class RouterProvider extends BaseProvider {
 
 		// First check instance models (populated by fetchModel)
 		if (this.models[id]) {
-			return { id, info: this.models[id] }
+			return { id, info: this.resolveModelInfo(this.models[id], this.models[id]) }
 		}
 
 		// Fall back to global cache (synchronous disk/memory cache).
@@ -110,14 +118,14 @@ export abstract class RouterProvider extends BaseProvider {
 		if (cachedModels?.[id]) {
 			// Also populate instance models for future calls
 			this.models = cachedModels
-			return { id, info: cachedModels[id] }
+			return { id, info: this.resolveModelInfo(cachedModels[id], cachedModels[id]) }
 		}
 
 		// Last resort: preserve the configured model ID (falling back to the default
 		// only when none is configured) so an as-yet-unfetched model isn't silently
 		// swapped for the hardcoded default. info still comes from defaults since we
 		// have no fetched or cached metadata for the configured model at this point.
-		return { id, info: this.defaultModelInfo }
+		return { id, info: this.resolveModelInfo(undefined, this.defaultModelInfo) }
 	}
 
 	protected supportsTemperature(modelId: string): boolean {
diff --git a/src/api/providers/unbound.ts b/src/api/providers/unbound.ts
index c3ec9c44fc..0ee069b1f5 100644
--- a/src/api/providers/unbound.ts
+++ b/src/api/providers/unbound.ts
@@ -1,7 +1,13 @@
 import { Anthropic } from "@anthropic-ai/sdk"
 import OpenAI from "openai"
 
-import { type ModelInfo, type ModelRecord, unboundDefaultModelId, unboundDefaultModelInfo } from "@roo-code/types"
+import {
+	applyCustomModelInfo,
+	type ModelInfo,
+	type ModelRecord,
+	unboundDefaultModelId,
+	unboundDefaultModelInfo,
+} from "@roo-code/types"
 
 import type { ApiHandlerOptions } from "../../shared/api"
 import { calculateApiCostOpenAI } from "../../shared/cost"
@@ -74,11 +80,13 @@ export class UnboundHandler extends BaseProvider implements SingleCompletionHand
 
 	override getModel() {
 		const id = this.options.unboundModelId ?? unboundDefaultModelId
-		const cachedInfo = this.models[id] ?? unboundDefaultModelInfo
+		const discoveredInfo = this.models[id]
+		const cachedInfo = discoveredInfo ?? unboundDefaultModelInfo
 		let info: ModelInfo = cachedInfo
 
 		// Apply tool preferences for models accessed through routers (OpenAI, Gemini)
 		info = applyRouterToolPreferences(id, info)
+		info = applyCustomModelInfo(discoveredInfo ? info : undefined, this.options) ?? info
 
 		const params = getModelParams({
 			format: "openai",
diff --git a/webview-ui/src/components/chat/TaskHeader.tsx b/webview-ui/src/components/chat/TaskHeader.tsx
index 0941a22e2b..4de87d9bdb 100644
--- a/webview-ui/src/components/chat/TaskHeader.tsx
+++ b/webview-ui/src/components/chat/TaskHeader.tsx
@@ -69,7 +69,8 @@ const TaskHeader = ({
 
 	const textContainerRef = useRef<HTMLDivElement>(null)
 	const textRef = useRef<HTMLDivElement>(null)
-	const contextWindow = model?.contextWindow || 1
+	const contextWindow = model?.contextWindow
+	const contextWindowForDisplay = typeof contextWindow === "number" && contextWindow > 0 ? contextWindow : undefined
 
 	// Calculate maxTokens (reserved for output) once for reuse in percentage and tooltip
 	const maxTokens = useMemo(
@@ -201,71 +202,82 @@ const TaskHeader = ({
 						</div>
 					</div>
 				</div>
-				{!isTaskExpanded && contextWindow > 0 && (
+				{!isTaskExpanded && (
 					<div
 						className="flex items-center justify-between text-sm text-muted-foreground/70"
 						onClick={(e) => e.stopPropagation()}>
 						<div className="flex items-center gap-2">
-							<StandardTooltip
-								content={(() => {
-									const availableSpace = contextWindow - (contextTokens || 0) - reservedForOutput
+							{contextWindowForDisplay !== undefined && (
+								<StandardTooltip
+									content={(() => {
+										const availableSpace =
+											contextWindowForDisplay - (contextTokens || 0) - reservedForOutput
 
-									return (
-										<Table className="text-base ml-1.5">
-											<TableBody>
-												<TableRow>
-													<TableCell className="font-medium whitespace-nowrap">
-														{t("chat:tokenProgress.tokensUsedLabel")}
-													</TableCell>
-													<TableCell className="text-right text-[0.9em] font-mono">
-														{formatLargeNumber(contextTokens || 0)} /{" "}
-														{formatLargeNumber(contextWindow)}
-													</TableCell>
-												</TableRow>
-												{reservedForOutput > 0 && (
-													<TableRow>
-														<TableCell className="font-medium whitespace-nowrap">
-															{t("chat:tokenProgress.reservedForResponseLabel")}
-														</TableCell>
-														<TableCell className="text-right text-[0.9em] font-mono">
-															{formatLargeNumber(reservedForOutput)}
-														</TableCell>
-													</TableRow>
-												)}
-												{availableSpace > 0 && (
+										return (
+											<Table className="text-base ml-1.5">
+												<TableBody>
 													<TableRow>
 														<TableCell className="font-medium whitespace-nowrap">
-															{t("chat:tokenProgress.availableSpaceLabel")}
+															{t("chat:tokenProgress.tokensUsedLabel")}
 														</TableCell>
 														<TableCell className="text-right text-[0.9em] font-mono">
-															{formatLargeNumber(availableSpace)}
+															{formatLargeNumber(contextTokens || 0)} /{" "}
+															{formatLargeNumber(contextWindowForDisplay)}
 														</TableCell>
 													</TableRow>
-												)}
-											</TableBody>
-										</Table>
-									)
-								})()}
-								side="top"
-								sideOffset={8}>
-								<span className="flex items-center gap-1.5">
-									{(() => {
-										// Calculate percentage of available input space used
-										// Available input space = context window - reserved for output
-										const availableInputSpace = contextWindow - reservedForOutput
-										const percentage =
-											availableInputSpace > 0
-												? Math.round(((contextTokens || 0) / availableInputSpace) * 100)
-												: 0
-										return (
-											<>
-												<CircularProgress percentage={percentage} />
-												<span>{percentage}%</span>
-											</>
+													{reservedForOutput > 0 && (
+														<TableRow>
+															<TableCell className="font-medium whitespace-nowrap">
+																{t("chat:tokenProgress.reservedForResponseLabel")}
+															</TableCell>
+															<TableCell className="text-right text-[0.9em] font-mono">
+																{formatLargeNumber(reservedForOutput)}
+															</TableCell>
+														</TableRow>
+													)}
+													{availableSpace > 0 && (
+														<TableRow>
+															<TableCell className="font-medium whitespace-nowrap">
+																{t("chat:tokenProgress.availableSpaceLabel")}
+															</TableCell>
+															<TableCell className="text-right text-[0.9em] font-mono">
+																{formatLargeNumber(availableSpace)}
+															</TableCell>
+														</TableRow>
+													)}
+												</TableBody>
+											</Table>
 										)
 									})()}
-								</span>
-							</StandardTooltip>
+									side="top"
+									sideOffset={8}>
+									<span className="flex items-center gap-1.5">
+										{(() => {
+											// Calculate percentage of available input space used
+											// Available input space = context window - reserved for output
+											const availableInputSpace = contextWindowForDisplay - reservedForOutput
+											const rawPercentage =
+												availableInputSpace > 0
+													? Math.round(((contextTokens || 0) / availableInputSpace) * 100)
+													: 0
+											const percentage = Math.min(100, rawPercentage)
+											const isAtLimit =
+												availableInputSpace > 0 && (contextTokens || 0) >= availableInputSpace
+											return (
+												<>
+													<CircularProgress percentage={percentage} />
+													<span
+														className={
+															isAtLimit ? "text-vscode-errorForeground" : undefined
+														}>
+														{percentage}%
+													</span>
+												</>
+											)
+										})()}
+									</span>
+								</StandardTooltip>
+							)}
 							{!!totalCost && (
 								<>
 									<span>·</span>
@@ -307,11 +319,13 @@ const TaskHeader = ({
 						<div
 							className="flex items-center gap-1 ml-8 w-60 min-w-[120px] shrink"
 							onClick={(e) => e.stopPropagation()}>
-							<ContextWindowProgress
-								contextWindow={contextWindow}
-								contextTokens={contextTokens || 0}
-								maxTokens={maxTokens ?? undefined}
-							/>
+							{contextWindowForDisplay !== undefined && (
+								<ContextWindowProgress
+									contextWindow={contextWindowForDisplay}
+									contextTokens={contextTokens || 0}
+									maxTokens={maxTokens ?? undefined}
+								/>
+							)}
 							{condenseButton}
 						</div>
 					</div>
@@ -342,7 +356,7 @@ const TaskHeader = ({
 						<div className="pt-3 mt-2 -mx-2.5 px-2.5 border-t border-vscode-sideBar-background">
 							<table className="w-full text-sm">
 								<tbody>
-									{contextWindow > 0 && (
+									{contextWindowForDisplay !== undefined ? (
 										<tr>
 											<th
 												className="font-medium text-left align-top w-1 whitespace-nowrap pr-3 h-[24px]"
@@ -350,9 +364,9 @@ const TaskHeader = ({
 												{t("chat:task.contextWindow")}
 											</th>
 											<td className="font-light align-top">
-												<div className={`max-w-md -mt-1.5 flex flex-nowrap gap-1`}>
+												<div className="max-w-md -mt-1.5 flex flex-nowrap gap-1">
 													<ContextWindowProgress
-														contextWindow={contextWindow}
+														contextWindow={contextWindowForDisplay}
 														contextTokens={contextTokens || 0}
 														maxTokens={maxTokens ?? undefined}
 													/>
@@ -360,6 +374,12 @@ const TaskHeader = ({
 												</div>
 											</td>
 										</tr>
+									) : (
+										<tr>
+											<td colSpan={2} className="font-light align-top">
+												<div className="flex justify-end">{condenseButton}</div>
+											</td>
+										</tr>
 									)}
 
 									<tr>
diff --git a/webview-ui/src/components/chat/__tests__/TaskHeader.spec.tsx b/webview-ui/src/components/chat/__tests__/TaskHeader.spec.tsx
index 9d23ca6886..7e5a9e3112 100644
--- a/webview-ui/src/components/chat/__tests__/TaskHeader.spec.tsx
+++ b/webview-ui/src/components/chat/__tests__/TaskHeader.spec.tsx
@@ -334,5 +334,27 @@ describe("TaskHeader", () => {
 
 			expect(screen.getByText("25%")).toBeInTheDocument()
 		})
+
+		it("should clamp over-limit usage to 100% and mark it as an error", () => {
+			renderTaskHeader({ contextTokens: 1000 })
+
+			const percentage = screen.getByText("100%")
+			expect(percentage).toHaveClass("text-vscode-errorForeground")
+			expect(screen.queryByText("125%")).not.toBeInTheDocument()
+		})
+
+		it("should keep the condense action available when context metadata is unavailable", () => {
+			mockModelInfo = undefined
+			mockMaxOutputTokens = 0
+
+			renderTaskHeader()
+
+			const condenseButton = screen
+				.getAllByRole("button")
+				.find((button) => button.querySelector("svg.lucide-list-chevrons-down-up"))
+
+			expect(condenseButton).toBeDefined()
+			expect(screen.queryByText(/%$/)).not.toBeInTheDocument()
+		})
 	})
 })
diff --git a/webview-ui/src/components/settings/ApiOptions.tsx b/webview-ui/src/components/settings/ApiOptions.tsx
index c5e69978ff..7c1dea3203 100644
--- a/webview-ui/src/components/settings/ApiOptions.tsx
+++ b/webview-ui/src/components/settings/ApiOptions.tsx
@@ -447,6 +447,7 @@ const ApiOptions = ({
 							setApiConfigurationField={setApiConfigurationField}
 							routerModels={routerModels}
 							selectedModelId={selectedModelId}
+							selectedModelInfo={selectedModelInfo}
 							uriScheme={uriScheme}
 							simplifySettings={fromWelcomeView}
 							organizationAllowList={organizationAllowList}
@@ -460,6 +461,7 @@ const ApiOptions = ({
 							apiConfiguration={apiConfiguration}
 							setApiConfigurationField={setApiConfigurationField}
 							routerModels={routerModels}
+							selectedModelInfo={selectedModelInfo}
 							refetchRouterModels={refetchRouterModels}
 							organizationAllowList={organizationAllowList}
 							modelValidationError={modelValidationError}
@@ -472,6 +474,7 @@ const ApiOptions = ({
 							apiConfiguration={apiConfiguration}
 							setApiConfigurationField={setApiConfigurationField}
 							routerModels={routerModels}
+							selectedModelInfo={selectedModelInfo}
 							refetchRouterModels={refetchRouterModels}
 							organizationAllowList={organizationAllowList}
 							modelValidationError={modelValidationError}
@@ -648,6 +651,7 @@ const ApiOptions = ({
 							apiConfiguration={apiConfiguration}
 							setApiConfigurationField={setApiConfigurationField}
 							routerModels={routerModels}
+							selectedModelInfo={selectedModelInfo}
 							organizationAllowList={organizationAllowList}
 							modelValidationError={modelValidationError}
 							simplifySettings={fromWelcomeView}
@@ -681,6 +685,7 @@ const ApiOptions = ({
 							apiConfiguration={apiConfiguration}
 							setApiConfigurationField={setApiConfigurationField}
 							routerModels={routerModels}
+							selectedModelInfo={selectedModelInfo}
 							organizationAllowList={organizationAllowList}
 							modelValidationError={modelValidationError}
 							simplifySettings={fromWelcomeView}
diff --git a/webview-ui/src/components/settings/CustomModelInfoSettings.tsx b/webview-ui/src/components/settings/CustomModelInfoSettings.tsx
new file mode 100644
index 0000000000..52fb15062f
--- /dev/null
+++ b/webview-ui/src/components/settings/CustomModelInfoSettings.tsx
@@ -0,0 +1,218 @@
+import { useEffect, useState } from "react"
+import { VSCodeCheckbox, VSCodeTextField } from "@vscode/webview-ui-toolkit/react"
+
+import { type CustomModelInfo, type ModelInfo, type ProviderSettings } from "@roo-code/types"
+
+import { Button, Collapsible, CollapsibleContent, CollapsibleTrigger } from "@src/components/ui"
+import { useAppTranslation } from "@src/i18n/TranslationContext"
+
+type CustomModelInfoSettingsProps = {
+	apiConfiguration: ProviderSettings
+	setApiConfigurationField: (field: "customModelInfo", value: ProviderSettings["customModelInfo"]) => void
+	selectedModelInfo?: ModelInfo
+}
+
+type ValueChangeEvent = {
+	target: EventTarget | null
+}
+
+const parsePositiveInteger = (value: string): number | undefined => {
+	const normalized = value.trim()
+
+	if (!/^\d+$/.test(normalized)) {
+		return undefined
+	}
+
+	const parsed = Number(normalized)
+	return Number.isSafeInteger(parsed) && parsed > 0 ? parsed : undefined
+}
+
+const getEventValue = (event: ValueChangeEvent): string => {
+	const target = event.target
+
+	if (target && "value" in target && typeof target.value === "string") {
+		return target.value
+	}
+
+	return ""
+}
+
+const getCheckboxValue = (event: ValueChangeEvent): boolean => {
+	const target = event.target
+
+	if (target && "checked" in target && typeof target.checked === "boolean") {
+		return target.checked
+	}
+
+	return false
+}
+
+const getInputBorderColor = (value: string): string | undefined => {
+	if (!value.trim()) {
+		return undefined
+	}
+
+	return parsePositiveInteger(value)
+		? "var(--vscode-testing-iconPassed)"
+		: "var(--vscode-inputValidation-errorBorder)"
+}
+
+export const CustomModelInfoSettings = ({
+	apiConfiguration,
+	setApiConfigurationField,
+	selectedModelInfo,
+}: CustomModelInfoSettingsProps) => {
+	const { t } = useAppTranslation()
+	const [isOpen, setIsOpen] = useState(!selectedModelInfo)
+	const [contextWindowInput, setContextWindowInput] = useState(
+		apiConfiguration.customModelInfo?.contextWindow?.toString() ?? "",
+	)
+	const [maxTokensInput, setMaxTokensInput] = useState(apiConfiguration.customModelInfo?.maxTokens?.toString() ?? "")
+
+	const configuredContextWindow = apiConfiguration.customModelInfo?.contextWindow
+	const configuredMaxTokens = apiConfiguration.customModelInfo?.maxTokens
+	const customModelInfo = apiConfiguration.customModelInfo ?? {}
+
+	useEffect(() => {
+		if (parsePositiveInteger(contextWindowInput) !== configuredContextWindow) {
+			setContextWindowInput(configuredContextWindow?.toString() ?? "")
+		}
+	}, [configuredContextWindow, contextWindowInput])
+
+	useEffect(() => {
+		if (parsePositiveInteger(maxTokensInput) !== configuredMaxTokens) {
+			setMaxTokensInput(configuredMaxTokens?.toString() ?? "")
+		}
+	}, [configuredMaxTokens, maxTokensInput])
+
+	useEffect(() => {
+		if (!selectedModelInfo) {
+			setIsOpen(true)
+		}
+	}, [selectedModelInfo])
+
+	const updateOverride = <K extends keyof CustomModelInfo>(field: K, value: CustomModelInfo[K] | undefined) => {
+		const next: CustomModelInfo = { ...customModelInfo }
+
+		if (value === undefined) {
+			delete next[field]
+		} else {
+			next[field] = value
+		}
+
+		setApiConfigurationField("customModelInfo", Object.keys(next).length > 0 ? next : undefined)
+	}
+
+	const handleContextWindowInput = (event: ValueChangeEvent) => {
+		const value = getEventValue(event)
+		setContextWindowInput(value)
+		updateOverride("contextWindow", parsePositiveInteger(value))
+	}
+
+	const handleMaxTokensInput = (event: ValueChangeEvent) => {
+		const value = getEventValue(event)
+		setMaxTokensInput(value)
+		updateOverride("maxTokens", parsePositiveInteger(value))
+	}
+
+	const resetOverrides = () => {
+		setContextWindowInput("")
+		setMaxTokensInput("")
+		setApiConfigurationField("customModelInfo", undefined)
+	}
+
+	const supportsImages = customModelInfo.supportsImages ?? selectedModelInfo?.supportsImages ?? false
+	const supportsPromptCache = customModelInfo.supportsPromptCache ?? selectedModelInfo?.supportsPromptCache ?? false
+	const contextWindowOverride = parsePositiveInteger(contextWindowInput)
+	const maxTokensOverride = parsePositiveInteger(maxTokensInput)
+	const hasInvalidContextWindow = contextWindowInput.trim().length > 0 && contextWindowOverride === undefined
+	const hasInvalidMaxTokens = maxTokensInput.trim().length > 0 && maxTokensOverride === undefined
+	const hasInvalidRange =
+		contextWindowOverride !== undefined &&
+		maxTokensOverride !== undefined &&
+		maxTokensOverride > contextWindowOverride
+
+	return (
+		<div className="mt-3 border-t border-vscode-panel-border pt-3">
+			<Collapsible open={isOpen} onOpenChange={setIsOpen}>
+				<CollapsibleTrigger className="flex w-full items-center gap-1 text-left text-sm font-medium hover:opacity-80">
+					<span className={`codicon codicon-chevron-${isOpen ? "down" : "right"}`} />
+					<span>{t("settings:providers.customModelInfo.title")}</span>
+				</CollapsibleTrigger>
+				<CollapsibleContent className="space-y-3 pt-2">
+					<p className="text-xs text-vscode-descriptionForeground">
+						{selectedModelInfo
+							? t("settings:providers.customModelInfo.description")
+							: t("settings:providers.customModelInfo.unresolved")}
+					</p>
+
+					<div className="grid grid-cols-1 gap-3 md:grid-cols-2">
+						<div className="flex flex-col gap-1">
+							<label htmlFor="custom-context-window" className="text-sm font-medium">
+								{t("settings:providers.customModelInfo.contextWindow.label")}
+							</label>
+							<VSCodeTextField
+								id="custom-context-window"
+								value={contextWindowInput}
+								onInput={handleContextWindowInput}
+								placeholder={selectedModelInfo?.contextWindow?.toString() ?? "0"}
+								style={{ borderColor: getInputBorderColor(contextWindowInput), width: "100%" }}
+								aria-invalid={hasInvalidContextWindow}
+							/>
+							<span className="text-xs text-vscode-descriptionForeground">
+								{t("settings:providers.customModelInfo.contextWindow.description")}
+							</span>
+						</div>
+
+						<div className="flex flex-col gap-1">
+							<label htmlFor="custom-max-tokens" className="text-sm font-medium">
+								{t("settings:providers.customModelInfo.maxTokens.label")}
+							</label>
+							<VSCodeTextField
+								id="custom-max-tokens"
+								value={maxTokensInput}
+								onInput={handleMaxTokensInput}
+								placeholder={selectedModelInfo?.maxTokens?.toString() ?? "0"}
+								style={{ borderColor: getInputBorderColor(maxTokensInput), width: "100%" }}
+								aria-invalid={hasInvalidMaxTokens}
+							/>
+							<span className="text-xs text-vscode-descriptionForeground">
+								{t("settings:providers.customModelInfo.maxTokens.description")}
+							</span>
+						</div>
+					</div>
+
+					{hasInvalidRange && (
+						<p className="text-xs text-vscode-errorForeground">
+							{t("settings:providers.customModelInfo.maxTokensWarning")}
+						</p>
+					)}
+
+					<div className="flex flex-col gap-2">
+						<VSCodeCheckbox
+							checked={supportsImages}
+							onChange={(event) => updateOverride("supportsImages", getCheckboxValue(event))}>
+							{t("settings:providers.customModelInfo.supportsImages.label")}
+						</VSCodeCheckbox>
+						<span className="-mt-1 pl-5 text-xs text-vscode-descriptionForeground">
+							{t("settings:providers.customModelInfo.supportsImages.description")}
+						</span>
+
+						<VSCodeCheckbox
+							checked={supportsPromptCache}
+							onChange={(event) => updateOverride("supportsPromptCache", getCheckboxValue(event))}>
+							{t("settings:providers.customModelInfo.supportsPromptCache.label")}
+						</VSCodeCheckbox>
+						<span className="-mt-1 pl-5 text-xs text-vscode-descriptionForeground">
+							{t("settings:providers.customModelInfo.supportsPromptCache.description")}
+						</span>
+					</div>
+
+					<Button type="button" variant="ghost" size="sm" onClick={resetOverrides} className="px-0">
+						{t("settings:providers.customModelInfo.reset")}
+					</Button>
+				</CollapsibleContent>
+			</Collapsible>
+		</div>
+	)
+}
diff --git a/webview-ui/src/components/settings/__tests__/CustomModelInfoSettings.spec.tsx b/webview-ui/src/components/settings/__tests__/CustomModelInfoSettings.spec.tsx
new file mode 100644
index 0000000000..36d9537772
--- /dev/null
+++ b/webview-ui/src/components/settings/__tests__/CustomModelInfoSettings.spec.tsx
@@ -0,0 +1,67 @@
+import { fireEvent, render, screen } from "@testing-library/react"
+
+import type { ModelInfo, ProviderSettings } from "@roo-code/types"
+
+import { CustomModelInfoSettings } from "../CustomModelInfoSettings"
+
+vi.mock("@src/i18n/TranslationContext", () => ({
+	useAppTranslation: () => ({
+		t: (key: string) => key,
+	}),
+}))
+
+describe("CustomModelInfoSettings", () => {
+	const modelInfo: ModelInfo = {
+		contextWindow: 128_000,
+		maxTokens: 16_384,
+		supportsImages: false,
+		supportsPromptCache: true,
+	}
+
+	it("keeps numeric edits in the cached provider configuration and supports reset", () => {
+		const setApiConfigurationField = vi.fn()
+		const apiConfiguration: ProviderSettings = {
+			apiProvider: "openrouter",
+			customModelInfo: { contextWindow: 64_000 },
+		}
+
+		render(
+			<CustomModelInfoSettings
+				apiConfiguration={apiConfiguration}
+				setApiConfigurationField={setApiConfigurationField}
+				selectedModelInfo={modelInfo}
+			/>,
+		)
+
+		fireEvent.click(screen.getByText("settings:providers.customModelInfo.title"))
+
+		const contextWindowInput = screen.getByLabelText("settings:providers.customModelInfo.contextWindow.label")
+		fireEvent.input(contextWindowInput, { target: { value: "128000" } })
+
+		expect(setApiConfigurationField).toHaveBeenLastCalledWith("customModelInfo", { contextWindow: 128_000 })
+
+		fireEvent.click(screen.getByText("settings:providers.customModelInfo.reset"))
+		expect(setApiConfigurationField).toHaveBeenLastCalledWith("customModelInfo", undefined)
+	})
+
+	it("keeps invalid numeric input visible without persisting it", () => {
+		const setApiConfigurationField = vi.fn()
+
+		render(
+			<CustomModelInfoSettings
+				apiConfiguration={{ apiProvider: "requesty" }}
+				setApiConfigurationField={setApiConfigurationField}
+				selectedModelInfo={modelInfo}
+			/>,
+		)
+
+		fireEvent.click(screen.getByText("settings:providers.customModelInfo.title"))
+
+		const maxTokensInput = screen.getByLabelText("settings:providers.customModelInfo.maxTokens.label")
+		fireEvent.input(maxTokensInput, { target: { value: "12abc" } })
+
+		expect(maxTokensInput).toHaveValue("12abc")
+		expect(maxTokensInput).toHaveAttribute("aria-invalid", "true")
+		expect(setApiConfigurationField).toHaveBeenLastCalledWith("customModelInfo", undefined)
+	})
+})
diff --git a/webview-ui/src/components/settings/providers/OpenRouter.tsx b/webview-ui/src/components/settings/providers/OpenRouter.tsx
index 2dba8c8459..8e7402f2be 100644
--- a/webview-ui/src/components/settings/providers/OpenRouter.tsx
+++ b/webview-ui/src/components/settings/providers/OpenRouter.tsx
@@ -4,6 +4,7 @@ import { VSCodeTextField } from "@vscode/webview-ui-toolkit/react"
 
 import {
 	type ProviderSettings,
+	type ModelInfo,
 	type OrganizationAllowList,
 	type RouterModels,
 	openRouterDefaultModelId,
@@ -16,6 +17,7 @@ import { VSCodeButtonLink } from "@src/components/common/VSCodeButtonLink"
 import { inputEventTransform } from "../transforms"
 
 import { ModelPicker } from "../ModelPicker"
+import { CustomModelInfoSettings } from "../CustomModelInfoSettings"
 import { OpenRouterBalanceDisplay } from "./OpenRouterBalanceDisplay"
 
 type OpenRouterProps = {
@@ -23,6 +25,7 @@ type OpenRouterProps = {
 	setApiConfigurationField: (field: keyof ProviderSettings, value: ProviderSettings[keyof ProviderSettings]) => void
 	routerModels?: RouterModels
 	selectedModelId: string
+	selectedModelInfo?: ModelInfo
 	uriScheme: string | undefined
 	simplifySettings?: boolean
 	organizationAllowList: OrganizationAllowList
@@ -37,6 +40,7 @@ export const OpenRouter = ({
 	simplifySettings,
 	organizationAllowList,
 	modelValidationError,
+	selectedModelInfo,
 }: OpenRouterProps) => {
 	const { t } = useAppTranslation()
 
@@ -115,6 +119,11 @@ export const OpenRouter = ({
 				errorMessage={modelValidationError}
 				simplifySettings={simplifySettings}
 			/>
+			<CustomModelInfoSettings
+				apiConfiguration={apiConfiguration}
+				setApiConfigurationField={setApiConfigurationField}
+				selectedModelInfo={selectedModelInfo}
+			/>
 		</>
 	)
 }
diff --git a/webview-ui/src/components/settings/providers/Requesty.tsx b/webview-ui/src/components/settings/providers/Requesty.tsx
index ba24a6aafb..7bc3718875 100644
--- a/webview-ui/src/components/settings/providers/Requesty.tsx
+++ b/webview-ui/src/components/settings/providers/Requesty.tsx
@@ -3,6 +3,7 @@ import { VSCodeCheckbox, VSCodeTextField } from "@vscode/webview-ui-toolkit/reac
 
 import {
 	type ProviderSettings,
+	type ModelInfo,
 	type OrganizationAllowList,
 	type RouterModels,
 	requestyDefaultModelId,
@@ -14,6 +15,7 @@ import { Button } from "@src/components/ui"
 
 import { inputEventTransform } from "../transforms"
 import { ModelPicker } from "../ModelPicker"
+import { CustomModelInfoSettings } from "../CustomModelInfoSettings"
 import { RequestyBalanceDisplay } from "./RequestyBalanceDisplay"
 import { getCallbackUrl } from "@/oauth/urls"
 import { toRequestyServiceUrl } from "@roo/utils/requesty"
@@ -22,6 +24,7 @@ type RequestyProps = {
 	apiConfiguration: ProviderSettings
 	setApiConfigurationField: (field: keyof ProviderSettings, value: ProviderSettings[keyof ProviderSettings]) => void
 	routerModels?: RouterModels
+	selectedModelInfo?: ModelInfo
 	refetchRouterModels: () => void
 	organizationAllowList: OrganizationAllowList
 	modelValidationError?: string
@@ -37,6 +40,7 @@ export const Requesty = ({
 	modelValidationError,
 	uriScheme,
 	simplifySettings,
+	selectedModelInfo,
 }: RequestyProps) => {
 	const { t } = useAppTranslation()
 
@@ -148,6 +152,11 @@ export const Requesty = ({
 				errorMessage={modelValidationError}
 				simplifySettings={simplifySettings}
 			/>
+			<CustomModelInfoSettings
+				apiConfiguration={apiConfiguration}
+				setApiConfigurationField={setApiConfigurationField}
+				selectedModelInfo={selectedModelInfo}
+			/>
 		</>
 	)
 }
diff --git a/webview-ui/src/components/settings/providers/Unbound.tsx b/webview-ui/src/components/settings/providers/Unbound.tsx
index 8c68241415..ae180fce05 100644
--- a/webview-ui/src/components/settings/providers/Unbound.tsx
+++ b/webview-ui/src/components/settings/providers/Unbound.tsx
@@ -3,6 +3,7 @@ import { VSCodeTextField } from "@vscode/webview-ui-toolkit/react"
 
 import {
 	type ProviderSettings,
+	type ModelInfo,
 	type OrganizationAllowList,
 	type RouterModels,
 	unboundDefaultModelId,
@@ -14,11 +15,13 @@ import { Button } from "@src/components/ui"
 
 import { inputEventTransform } from "../transforms"
 import { ModelPicker } from "../ModelPicker"
+import { CustomModelInfoSettings } from "../CustomModelInfoSettings"
 
 type UnboundProps = {
 	apiConfiguration: ProviderSettings
 	setApiConfigurationField: (field: keyof ProviderSettings, value: ProviderSettings[keyof ProviderSettings]) => void
 	routerModels?: RouterModels
+	selectedModelInfo?: ModelInfo
 	refetchRouterModels: () => void
 	organizationAllowList: OrganizationAllowList
 	modelValidationError?: string
@@ -32,6 +35,7 @@ export const Unbound = ({
 	organizationAllowList,
 	modelValidationError,
 	simplifySettings,
+	selectedModelInfo,
 }: UnboundProps) => {
 	const { t } = useAppTranslation()
 
@@ -96,6 +100,11 @@ export const Unbound = ({
 				errorMessage={modelValidationError}
 				simplifySettings={simplifySettings}
 			/>
+			<CustomModelInfoSettings
+				apiConfiguration={apiConfiguration}
+				setApiConfigurationField={setApiConfigurationField}
+				selectedModelInfo={selectedModelInfo}
+			/>
 		</>
 	)
 }
diff --git a/webview-ui/src/components/settings/providers/VercelAiGateway.tsx b/webview-ui/src/components/settings/providers/VercelAiGateway.tsx
index 1f003ed52b..b83355f8e9 100644
--- a/webview-ui/src/components/settings/providers/VercelAiGateway.tsx
+++ b/webview-ui/src/components/settings/providers/VercelAiGateway.tsx
@@ -3,6 +3,7 @@ import { VSCodeTextField } from "@vscode/webview-ui-toolkit/react"
 
 import {
 	type ProviderSettings,
+	type ModelInfo,
 	type OrganizationAllowList,
 	type RouterModels,
 	vercelAiGatewayDefaultModelId,
@@ -13,11 +14,13 @@ import { VSCodeButtonLink } from "@src/components/common/VSCodeButtonLink"
 
 import { inputEventTransform } from "../transforms"
 import { ModelPicker } from "../ModelPicker"
+import { CustomModelInfoSettings } from "../CustomModelInfoSettings"
 
 type VercelAiGatewayProps = {
 	apiConfiguration: ProviderSettings
 	setApiConfigurationField: (field: keyof ProviderSettings, value: ProviderSettings[keyof ProviderSettings]) => void
 	routerModels?: RouterModels
+	selectedModelInfo?: ModelInfo
 	organizationAllowList: OrganizationAllowList
 	modelValidationError?: string
 	simplifySettings?: boolean
@@ -30,6 +33,7 @@ export const VercelAiGateway = ({
 	organizationAllowList,
 	modelValidationError,
 	simplifySettings,
+	selectedModelInfo,
 }: VercelAiGatewayProps) => {
 	const { t } = useAppTranslation()
 
@@ -77,6 +81,11 @@ export const VercelAiGateway = ({
 				errorMessage={modelValidationError}
 				simplifySettings={simplifySettings}
 			/>
+			<CustomModelInfoSettings
+				apiConfiguration={apiConfiguration}
+				setApiConfigurationField={setApiConfigurationField}
+				selectedModelInfo={selectedModelInfo}
+			/>
 		</>
 	)
 }
diff --git a/webview-ui/src/components/settings/providers/ZooGateway.tsx b/webview-ui/src/components/settings/providers/ZooGateway.tsx
index ac99f464a4..87e7290d65 100644
--- a/webview-ui/src/components/settings/providers/ZooGateway.tsx
+++ b/webview-ui/src/components/settings/providers/ZooGateway.tsx
@@ -1,6 +1,7 @@
 import { useEffect, useMemo } from "react"
 import {
 	type ProviderSettings,
+	type ModelInfo,
 	type OrganizationAllowList,
 	type RouterModels,
 	zooGatewayDefaultModelId,
@@ -12,12 +13,14 @@ import { useAppTranslation } from "@src/i18n/TranslationContext"
 import { VSCodeButtonLink } from "@src/components/common/VSCodeButtonLink"
 
 import { ModelPicker } from "../ModelPicker"
+import { CustomModelInfoSettings } from "../CustomModelInfoSettings"
 import { ApiErrorMessage } from "../ApiErrorMessage"
 
 type ZooGatewayProps = {
 	apiConfiguration: ProviderSettings
 	setApiConfigurationField: (field: keyof ProviderSettings, value: ProviderSettings[keyof ProviderSettings]) => void
 	routerModels?: RouterModels
+	selectedModelInfo?: ModelInfo
 	organizationAllowList: OrganizationAllowList
 	modelValidationError?: string
 	simplifySettings?: boolean
@@ -55,6 +58,7 @@ export const ZooGateway = ({
 	organizationAllowList,
 	modelValidationError,
 	simplifySettings,
+	selectedModelInfo,
 }: ZooGatewayProps) => {
 	const { t } = useAppTranslation()
 	const { zooCodeIsAuthenticated, zooCodeUserEmail, zooCodeUserName, zooCodeBaseUrl, uriScheme, deviceName } =
@@ -73,7 +77,7 @@ export const ZooGateway = ({
 		}
 
 		const current = apiConfiguration.zooGatewayModelId
-		if (!current || !modelIds.includes(current)) {
+		if (!current) {
 			setApiConfigurationField("zooGatewayModelId", resolvedDefaultModelId)
 		}
 	}, [apiConfiguration.zooGatewayModelId, modelIds, resolvedDefaultModelId, setApiConfigurationField])
@@ -120,6 +124,11 @@ export const ZooGateway = ({
 				errorMessage={modelValidationError}
 				simplifySettings={simplifySettings}
 			/>
+			<CustomModelInfoSettings
+				apiConfiguration={apiConfiguration}
+				setApiConfigurationField={setApiConfigurationField}
+				selectedModelInfo={selectedModelInfo}
+			/>
 		</>
 	)
 }
diff --git a/webview-ui/src/components/settings/providers/__tests__/ZooGateway.spec.tsx b/webview-ui/src/components/settings/providers/__tests__/ZooGateway.spec.tsx
index 9bdda1f433..aabf8a4796 100644
--- a/webview-ui/src/components/settings/providers/__tests__/ZooGateway.spec.tsx
+++ b/webview-ui/src/components/settings/providers/__tests__/ZooGateway.spec.tsx
@@ -106,7 +106,7 @@ describe("ZooGateway component", () => {
 		})
 	})
 
-	it("reassigns a stale model id that is not in the catalog", async () => {
+	it("preserves a configured model id that is not in the catalog", async () => {
 		const setApiConfigurationField = vi.fn()
 		render(
 			<ZooGateway
@@ -126,10 +126,7 @@ describe("ZooGateway component", () => {
 		)
 
 		await waitFor(() => {
-			expect(setApiConfigurationField).toHaveBeenCalledWith(
-				"zooGatewayModelId",
-				"anthropic.claude-sonnet-4-5-20250929-v1:0",
-			)
+			expect(setApiConfigurationField).not.toHaveBeenCalled()
 		})
 	})
 
diff --git a/webview-ui/src/components/ui/hooks/__tests__/useSelectedModel.spec.ts b/webview-ui/src/components/ui/hooks/__tests__/useSelectedModel.spec.ts
index 5fca23ba8e..e1af24a3c7 100644
--- a/webview-ui/src/components/ui/hooks/__tests__/useSelectedModel.spec.ts
+++ b/webview-ui/src/components/ui/hooks/__tests__/useSelectedModel.spec.ts
@@ -110,7 +110,7 @@ describe("useSelectedModel", () => {
 			})
 		})
 
-		it("should fall back to default when configured model doesn't exist in available models", () => {
+		it("should preserve a configured model when it is absent from available models", () => {
 			const specificProviderInfo: ModelInfo = {
 				maxTokens: 8192,
 				contextWindow: 16384,
@@ -159,22 +159,8 @@ describe("useSelectedModel", () => {
 			const wrapper = createWrapper()
 			const { result } = renderHook(() => useSelectedModel(apiConfiguration), { wrapper })
 
-			// Should fall back to provider default since "test-model" doesn't exist
-			expect(result.current.id).toBe("anthropic/claude-sonnet-4.5")
-			// Should still use specific provider info for the default model if specified
-			expect(result.current.info).toEqual({
-				...{
-					maxTokens: 8192,
-					contextWindow: 200_000,
-					supportsImages: true,
-					supportsPromptCache: true,
-					inputPrice: 3.0,
-					outputPrice: 15.0,
-					cacheWritesPrice: 3.75,
-					cacheReadsPrice: 0.3,
-				},
-				...specificProviderInfo,
-			})
+			expect(result.current.id).toBe("test-model")
+			expect(result.current.info).toEqual(specificProviderInfo)
 		})
 
 		it("should demonstrate the merging behavior validates the comment about missing fields", () => {
@@ -277,7 +263,7 @@ describe("useSelectedModel", () => {
 			expect(result.current.info).toEqual(baseModelInfo)
 		})
 
-		it("should fall back to default when configured model and provider don't exist", () => {
+		it("should preserve an unknown configured model when its provider metadata is unavailable", () => {
 			mockUseRouterModels.mockReturnValue({
 				data: {
 					openrouter: {
@@ -315,23 +301,46 @@ describe("useSelectedModel", () => {
 			const wrapper = createWrapper()
 			const { result } = renderHook(() => useSelectedModel(apiConfiguration), { wrapper })
 
-			// Should fall back to provider default since "non-existent-model" doesn't exist
-			expect(result.current.id).toBe("anthropic/claude-sonnet-4.5")
-			// Should use base model info since provider doesn't exist
-			expect(result.current.info).toEqual({
-				maxTokens: 8192,
-				contextWindow: 200_000,
-				supportsImages: true,
-				supportsPromptCache: true,
-				inputPrice: 3.0,
-				outputPrice: 15.0,
-				cacheWritesPrice: 3.75,
-				cacheReadsPrice: 0.3,
-			})
+			expect(result.current.id).toBe("non-existent-model")
+			expect(result.current.info).toBeUndefined()
 		})
 	})
 
 	describe("loading and error states", () => {
+		it("preserves a router model ID and applies custom metadata while model data is loading", () => {
+			mockUseRouterModels.mockReturnValue({
+				data: undefined,
+				isLoading: true,
+				isError: false,
+			} as any)
+
+			mockUseOpenRouterModelProviders.mockReturnValue({
+				data: undefined,
+				isLoading: false,
+				isError: false,
+			} as any)
+
+			const apiConfiguration: ProviderSettings = {
+				apiProvider: providerIdentifiers.openrouter,
+				openRouterModelId: "provider/future-model",
+				customModelInfo: {
+					contextWindow: 100_000,
+					maxTokens: 10_000,
+					supportsImages: true,
+				},
+			}
+
+			const wrapper = createWrapper()
+			const { result } = renderHook(() => useSelectedModel(apiConfiguration), { wrapper })
+
+			expect(result.current.id).toBe("provider/future-model")
+			expect(result.current.info).toMatchObject({
+				contextWindow: 100_000,
+				maxTokens: 10_000,
+				supportsImages: true,
+			})
+		})
+
 		it("should set loading when router models are loading for the default OpenRouter provider", () => {
 			mockUseRouterModels.mockReturnValue({
 				data: undefined,
diff --git a/webview-ui/src/components/ui/hooks/useSelectedModel.ts b/webview-ui/src/components/ui/hooks/useSelectedModel.ts
index ec513ce885..6a0f503850 100644
--- a/webview-ui/src/components/ui/hooks/useSelectedModel.ts
+++ b/webview-ui/src/components/ui/hooks/useSelectedModel.ts
@@ -24,6 +24,7 @@ import {
 	mainlandZAiModels,
 	fireworksModels,
 	friendliModels,
+	applyCustomModelInfo,
 	basetenModels,
 	qwenCodeModels,
 	kimiCodeDefaultModelInfo,
@@ -45,17 +46,50 @@ import { useLmStudioModels } from "./useLmStudioModels"
 import { useOllamaModels } from "./useOllamaModels"
 
 /**
- * Helper to get a validated model ID for dynamic providers.
- * Returns the configured model ID if it exists in the available models, otherwise returns the default.
+ * Helper to get a model ID for dynamic providers.
+ * Some router providers accept arbitrary model IDs, so their configured value
+ * must survive both an empty list and a list that does not contain the ID.
  */
 function getValidatedModelId(
 	configuredId: string | undefined,
 	availableModels: ModelRecord | undefined,
 	defaultModelId: string,
+	preserveConfiguredId = false,
 ): string {
+	if (preserveConfiguredId && configuredId) {
+		return configuredId
+	}
+
 	return configuredId && availableModels?.[configuredId] ? configuredId : defaultModelId
 }
 
+function getConfiguredRouterModelId(provider: ProviderName, apiConfiguration: ProviderSettings): string | undefined {
+	switch (provider) {
+		case providerIdentifiers.openrouter:
+			return apiConfiguration.openRouterModelId
+		case providerIdentifiers.requesty:
+			return apiConfiguration.requestyModelId
+		case providerIdentifiers.unbound:
+			return apiConfiguration.unboundModelId
+		case providerIdentifiers.vercelAiGateway:
+			return apiConfiguration.vercelAiGatewayModelId
+		case providerIdentifiers.zooGateway:
+			return apiConfiguration.zooGatewayModelId
+		default:
+			return undefined
+	}
+}
+
+function supportsCustomModelInfo(provider: ProviderName): boolean {
+	return (
+		provider === providerIdentifiers.openrouter ||
+		provider === providerIdentifiers.requesty ||
+		provider === providerIdentifiers.unbound ||
+		provider === providerIdentifiers.vercelAiGateway ||
+		provider === providerIdentifiers.zooGateway
+	)
+}
+
 export const useSelectedModel = (apiConfiguration?: ProviderSettings) => {
 	const provider = apiConfiguration?.apiProvider || "openrouter"
 	const activeProvider: ProviderName | undefined = isRetiredProvider(provider) ? undefined : provider
@@ -95,7 +129,7 @@ export const useSelectedModel = (apiConfiguration?: ProviderSettings) => {
 		hasValidRouterData &&
 		(!needOpenRouterProviders || typeof openRouterModelProviders.data !== "undefined")
 
-	const { id, info } =
+	const selectedModel =
 		apiConfiguration && isReady && activeProvider
 			? getSelectedModel({
 					provider: activeProvider,
@@ -110,7 +144,20 @@ export const useSelectedModel = (apiConfiguration?: ProviderSettings) => {
 						id: apiConfiguration.apiModelId || getProviderDefaultModelId("kimi-code"),
 						info: kimiCodeDefaultModelInfo,
 					}
-				: { id: getProviderDefaultModelId(activeProvider ?? "openrouter"), info: undefined }
+				: {
+						id:
+							(activeProvider &&
+								apiConfiguration &&
+								getConfiguredRouterModelId(activeProvider, apiConfiguration)) ||
+							getProviderDefaultModelId(activeProvider ?? "openrouter"),
+						info: undefined,
+					}
+
+	const { id } = selectedModel
+	const info =
+		activeProvider && supportsCustomModelInfo(activeProvider)
+			? applyCustomModelInfo(selectedModel.info, apiConfiguration)
+			: selectedModel.info
 
 	return {
 		provider,
@@ -150,7 +197,12 @@ function getSelectedModel({
 	const defaultModelId = getProviderDefaultModelId(provider)
 	switch (provider) {
 		case providerIdentifiers.openrouter: {
-			const id = getValidatedModelId(apiConfiguration.openRouterModelId, routerModels.openrouter, defaultModelId)
+			const id = getValidatedModelId(
+				apiConfiguration.openRouterModelId,
+				routerModels.openrouter,
+				defaultModelId,
+				true,
+			)
 			let info = routerModels.openrouter?.[id]
 			const specificProvider = apiConfiguration.openRouterSpecificProvider
 
@@ -166,12 +218,17 @@ function getSelectedModel({
 			return { id, info }
 		}
 		case providerIdentifiers.requesty: {
-			const id = getValidatedModelId(apiConfiguration.requestyModelId, routerModels.requesty, defaultModelId)
+			const id = getValidatedModelId(
+				apiConfiguration.requestyModelId,
+				routerModels.requesty,
+				defaultModelId,
+				true,
+			)
 			const routerInfo = routerModels.requesty?.[id]
 			return { id, info: routerInfo }
 		}
 		case providerIdentifiers.unbound: {
-			const id = getValidatedModelId(apiConfiguration.unboundModelId, routerModels.unbound, defaultModelId)
+			const id = getValidatedModelId(apiConfiguration.unboundModelId, routerModels.unbound, defaultModelId, true)
 			const routerInfo = routerModels.unbound?.[id]
 			return { id, info: routerInfo }
 		}
@@ -387,6 +444,7 @@ function getSelectedModel({
 				apiConfiguration.vercelAiGatewayModelId,
 				routerModels["vercel-ai-gateway"],
 				defaultModelId,
+				true,
 			)
 			const info = routerModels["vercel-ai-gateway"]?.[id]
 			return { id, info }
@@ -414,6 +472,7 @@ function getSelectedModel({
 				apiConfiguration.zooGatewayModelId,
 				routerModels["zoo-gateway"],
 				defaultModelId,
+				true,
 			)
 			const info = routerModels["zoo-gateway"]?.[id]
 			return { id, info }
diff --git a/webview-ui/src/i18n/locales/en/settings.json b/webview-ui/src/i18n/locales/en/settings.json
index 3d84065849..49b2d61e10 100644
--- a/webview-ui/src/i18n/locales/en/settings.json
+++ b/webview-ui/src/i18n/locales/en/settings.json
@@ -697,6 +697,29 @@
 			},
 			"resetDefaults": "Reset to Defaults"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Rate limit",
 			"description": "Minimum time between API requests."

From e382bfe3482226f5f694d3eaaeffd990f1793d64 Mon Sep 17 00:00:00 2001
From: everyoneexe <mansurozdemir41@gmail.com>
Date: Tue, 4 Aug 2026 21:40:07 +0200
Subject: [PATCH 3/3] fix: address custom model metadata review feedback

---
 .../src/__tests__/custom-model-info.test.ts   |   7 +
 packages/types/src/model.ts                   |  10 +-
 src/api/providers/__tests__/kenari.spec.ts    |  11 +
 .../providers/__tests__/openrouter.spec.ts    |  46 ++++
 src/api/providers/__tests__/requesty.spec.ts  |  19 ++
 src/api/providers/__tests__/unbound.spec.ts   |  19 ++
 .../__tests__/vercel-ai-gateway.spec.ts       |  12 +
 .../providers/__tests__/zoo-gateway.spec.ts   |  12 +
 webview-ui/src/components/chat/TaskHeader.tsx |   5 +-
 .../chat/__tests__/TaskHeader.spec.tsx        |  10 +
 .../settings/CustomModelInfoSettings.tsx      |  12 +-
 .../CustomModelInfoSettings.spec.tsx          |  57 +++++
 .../hooks/__tests__/useSelectedModel.spec.ts  | 227 +++++++++++++++++-
 webview-ui/src/i18n/locales/ca/settings.json  |  23 ++
 webview-ui/src/i18n/locales/de/settings.json  |  23 ++
 webview-ui/src/i18n/locales/es/settings.json  |  23 ++
 webview-ui/src/i18n/locales/fr/settings.json  |  23 ++
 webview-ui/src/i18n/locales/hi/settings.json  |  23 ++
 webview-ui/src/i18n/locales/id/settings.json  |  23 ++
 webview-ui/src/i18n/locales/it/settings.json  |  23 ++
 webview-ui/src/i18n/locales/ja/settings.json  |  23 ++
 webview-ui/src/i18n/locales/ko/settings.json  |  23 ++
 webview-ui/src/i18n/locales/nl/settings.json  |  23 ++
 webview-ui/src/i18n/locales/pl/settings.json  |  23 ++
 .../src/i18n/locales/pt-BR/settings.json      |  23 ++
 webview-ui/src/i18n/locales/ru/settings.json  |  23 ++
 webview-ui/src/i18n/locales/tr/settings.json  |  23 ++
 webview-ui/src/i18n/locales/vi/settings.json  |  23 ++
 .../src/i18n/locales/zh-CN/settings.json      |  23 ++
 .../src/i18n/locales/zh-TW/settings.json      |  23 ++
 30 files changed, 814 insertions(+), 24 deletions(-)

diff --git a/packages/types/src/__tests__/custom-model-info.test.ts b/packages/types/src/__tests__/custom-model-info.test.ts
index 1d2b3da830..a05a58e42c 100644
--- a/packages/types/src/__tests__/custom-model-info.test.ts
+++ b/packages/types/src/__tests__/custom-model-info.test.ts
@@ -65,4 +65,11 @@ describe("custom model info", () => {
 			}).success,
 		).toBe(false)
 	})
+
+	it("rejects unsafe integer overrides", () => {
+		const unsafeInteger = Number.MAX_SAFE_INTEGER + 1
+
+		expect(customModelInfoSchema.safeParse({ contextWindow: unsafeInteger }).success).toBe(false)
+		expect(customModelInfoSchema.safeParse({ maxTokens: unsafeInteger }).success).toBe(false)
+	})
 })
diff --git a/packages/types/src/model.ts b/packages/types/src/model.ts
index b7c74219c0..9784ae0ba3 100644
--- a/packages/types/src/model.ts
+++ b/packages/types/src/model.ts
@@ -188,10 +188,16 @@ export type ModelInfo = z.infer<typeof modelInfoSchema>
  * or unavailable. This is intentionally narrower than ModelInfo: prices and
  * other accounting fields must remain provider-owned.
  */
+const positiveSafeIntegerSchema = z
+	.number()
+	.int()
+	.positive()
+	.refine(Number.isSafeInteger, { message: "Expected a safe integer" })
+
 export const customModelInfoSchema = z
 	.object({
-		maxTokens: z.number().int().positive().optional(),
-		contextWindow: z.number().int().positive().optional(),
+		maxTokens: positiveSafeIntegerSchema.optional(),
+		contextWindow: positiveSafeIntegerSchema.optional(),
 		supportsImages: z.boolean().optional(),
 		supportsPromptCache: z.boolean().optional(),
 	})
diff --git a/src/api/providers/__tests__/kenari.spec.ts b/src/api/providers/__tests__/kenari.spec.ts
index d6b95ce0b1..bdbb1d2000 100644
--- a/src/api/providers/__tests__/kenari.spec.ts
+++ b/src/api/providers/__tests__/kenari.spec.ts
@@ -76,6 +76,17 @@ describe("KenariHandler", () => {
 			expect(result.info.supportsPromptCache).toBe(false)
 		})
 
+		it("does not apply gateway-only custom metadata overrides", async () => {
+			const handler = new KenariHandler({
+				...mockOptions,
+				customModelInfo: { contextWindow: 100_000, maxTokens: 10_000 },
+			})
+			const result = await handler.fetchModel()
+
+			expect(result.info.contextWindow).toBe(1_048_576)
+			expect(result.info.maxTokens).toBe(32_768)
+		})
+
 		it("falls back to the default model id when none is configured", async () => {
 			const handler = new KenariHandler({ kenariApiKey: "test-key" })
 			const result = await handler.fetchModel()
diff --git a/src/api/providers/__tests__/openrouter.spec.ts b/src/api/providers/__tests__/openrouter.spec.ts
index 6f5d42ab10..5721b45872 100644
--- a/src/api/providers/__tests__/openrouter.spec.ts
+++ b/src/api/providers/__tests__/openrouter.spec.ts
@@ -18,6 +18,7 @@ import { Anthropic } from "@anthropic-ai/sdk"
 import OpenAI from "openai"
 
 import { OpenRouterHandler } from "../openrouter"
+import { getModelEndpoints } from "../fetchers/modelEndpointCache"
 import { ApiHandlerOptions } from "../../../shared/api"
 import { Package } from "../../../shared/package"
 import { asyncStreamFrom, collectStream } from "../../../test-utils/stream"
@@ -101,6 +102,10 @@ vitest.mock("../fetchers/modelCache", () => ({
 	}),
 }))
 
+vitest.mock("../fetchers/modelEndpointCache", () => ({
+	getModelEndpoints: vitest.fn().mockResolvedValue({}),
+}))
+
 describe("OpenRouterHandler", () => {
 	const mockOptions: ApiHandlerOptions = {
 		openRouterApiKey: "test-key",
@@ -159,6 +164,47 @@ describe("OpenRouterHandler", () => {
 			expect(result.maxTokens).toBe(10_000)
 		})
 
+		it("applies custom metadata to a discovered specific-provider endpoint", async () => {
+			vitest.mocked(getModelEndpoints).mockResolvedValue({
+				"test-provider": {
+					contextWindow: 128_000,
+					maxTokens: 16_384,
+					supportsImages: true,
+					supportsPromptCache: true,
+				},
+			})
+
+			const handler = new OpenRouterHandler({
+				...mockOptions,
+				openRouterSpecificProvider: "test-provider",
+				customModelInfo: { contextWindow: 100_000, maxTokens: 10_000 },
+			})
+
+			const result = await handler.fetchModel()
+
+			expect(result.info.contextWindow).toBe(100_000)
+			expect(result.info.maxTokens).toBe(10_000)
+		})
+
+		it("synthesizes metadata for an unlisted configured model", async () => {
+			const modelId = "provider/unlisted-model"
+			const handler = new OpenRouterHandler({
+				...mockOptions,
+				openRouterModelId: modelId,
+				customModelInfo: {
+					contextWindow: 100_000,
+					maxTokens: 10_000,
+				},
+			})
+
+			const result = await handler.fetchModel()
+
+			expect(result.id).toBe(modelId)
+			expect(result.info.contextWindow).toBe(100_000)
+			expect(result.info.maxTokens).toBe(10_000)
+			expect(result.maxTokens).toBe(10_000)
+		})
+
 		it("returns default model info when options are not provided", async () => {
 			const handler = new OpenRouterHandler({})
 			const result = await handler.fetchModel()
diff --git a/src/api/providers/__tests__/requesty.spec.ts b/src/api/providers/__tests__/requesty.spec.ts
index 8d0d203d1d..0cb2b017ee 100644
--- a/src/api/providers/__tests__/requesty.spec.ts
+++ b/src/api/providers/__tests__/requesty.spec.ts
@@ -176,6 +176,25 @@ describe("RequestyHandler", () => {
 			expect(result.maxTokens).toBe(10_000)
 		})
 
+		it("synthesizes metadata for an unlisted configured model", async () => {
+			const modelId = "provider/unlisted-model"
+			const handler = new RequestyHandler({
+				...mockOptions,
+				requestyModelId: modelId,
+				customModelInfo: {
+					contextWindow: 100_000,
+					maxTokens: 10_000,
+				},
+			})
+
+			const result = await handler.fetchModel()
+
+			expect(result.id).toBe(modelId)
+			expect(result.info.contextWindow).toBe(100_000)
+			expect(result.info.maxTokens).toBe(10_000)
+			expect(result.maxTokens).toBe(10_000)
+		})
+
 		it("returns default model info when options are not provided", async () => {
 			const handler = new RequestyHandler({})
 			const result = await handler.fetchModel()
diff --git a/src/api/providers/__tests__/unbound.spec.ts b/src/api/providers/__tests__/unbound.spec.ts
index b9d7f8acb2..7af68942a3 100644
--- a/src/api/providers/__tests__/unbound.spec.ts
+++ b/src/api/providers/__tests__/unbound.spec.ts
@@ -59,6 +59,25 @@ describe("UnboundHandler", () => {
 		expect(result.maxTokens).toBe(10_000)
 	})
 
+	it("synthesizes metadata for an unlisted configured model", async () => {
+		const modelId = "provider/unlisted-model"
+		const handler = new UnboundHandler({
+			unboundApiKey: "test-key",
+			unboundModelId: modelId,
+			customModelInfo: {
+				contextWindow: 100_000,
+				maxTokens: 10_000,
+			},
+		})
+
+		const result = await handler.fetchModel()
+
+		expect(result.id).toBe(modelId)
+		expect(result.info.contextWindow).toBe(100_000)
+		expect(result.info.maxTokens).toBe(10_000)
+		expect(result.maxTokens).toBe(10_000)
+	})
+
 	it("identifies itself as Zoo Code in the Unbound request headers", () => {
 		new UnboundHandler({
 			unboundApiKey: "test-key",
diff --git a/src/api/providers/__tests__/vercel-ai-gateway.spec.ts b/src/api/providers/__tests__/vercel-ai-gateway.spec.ts
index 92cc785951..999be4855e 100644
--- a/src/api/providers/__tests__/vercel-ai-gateway.spec.ts
+++ b/src/api/providers/__tests__/vercel-ai-gateway.spec.ts
@@ -165,6 +165,18 @@ describe("VercelAiGatewayHandler", () => {
 			expect(result.info.supportsPromptCache).toBe(true)
 		})
 
+		it("applies custom metadata overrides to the discovered model", async () => {
+			const handler = new VercelAiGatewayHandler({
+				...mockOptions,
+				customModelInfo: { contextWindow: 100_000, maxTokens: 10_000, supportsImages: false },
+			})
+			const result = await handler.fetchModel()
+
+			expect(result.info.contextWindow).toBe(100_000)
+			expect(result.info.maxTokens).toBe(10_000)
+			expect(result.info.supportsImages).toBe(false)
+		})
+
 		it("returns default model info when options are not provided", async () => {
 			const handler = new VercelAiGatewayHandler({})
 			const result = await handler.fetchModel()
diff --git a/src/api/providers/__tests__/zoo-gateway.spec.ts b/src/api/providers/__tests__/zoo-gateway.spec.ts
index e797dc9745..254f76d2c4 100644
--- a/src/api/providers/__tests__/zoo-gateway.spec.ts
+++ b/src/api/providers/__tests__/zoo-gateway.spec.ts
@@ -214,6 +214,18 @@ describe("ZooGatewayHandler", () => {
 			expect(result.info.supportsPromptCache).toBe(true)
 		})
 
+		it("applies custom metadata overrides to the discovered model", async () => {
+			const handler = new ZooGatewayHandler({
+				...mockOptions,
+				customModelInfo: { contextWindow: 100_000, maxTokens: 10_000, supportsPromptCache: false },
+			})
+			const result = await handler.fetchModel()
+
+			expect(result.info.contextWindow).toBe(100_000)
+			expect(result.info.maxTokens).toBe(10_000)
+			expect(result.info.supportsPromptCache).toBe(false)
+		})
+
 		it("falls back to the default model when none is configured", async () => {
 			const handler = new ZooGatewayHandler({ zooSessionToken: "zoo_ext_test_token" })
 			const result = await handler.fetchModel()
diff --git a/webview-ui/src/components/chat/TaskHeader.tsx b/webview-ui/src/components/chat/TaskHeader.tsx
index 4de87d9bdb..bb070a71cd 100644
--- a/webview-ui/src/components/chat/TaskHeader.tsx
+++ b/webview-ui/src/components/chat/TaskHeader.tsx
@@ -70,7 +70,10 @@ const TaskHeader = ({
 	const textContainerRef = useRef<HTMLDivElement>(null)
 	const textRef = useRef<HTMLDivElement>(null)
 	const contextWindow = model?.contextWindow
-	const contextWindowForDisplay = typeof contextWindow === "number" && contextWindow > 0 ? contextWindow : undefined
+	const contextWindowForDisplay =
+		typeof contextWindow === "number" && Number.isFinite(contextWindow) && contextWindow > 0
+			? contextWindow
+			: undefined
 
 	// Calculate maxTokens (reserved for output) once for reuse in percentage and tooltip
 	const maxTokens = useMemo(
diff --git a/webview-ui/src/components/chat/__tests__/TaskHeader.spec.tsx b/webview-ui/src/components/chat/__tests__/TaskHeader.spec.tsx
index 7e5a9e3112..9d7cbf5e08 100644
--- a/webview-ui/src/components/chat/__tests__/TaskHeader.spec.tsx
+++ b/webview-ui/src/components/chat/__tests__/TaskHeader.spec.tsx
@@ -356,5 +356,15 @@ describe("TaskHeader", () => {
 			expect(condenseButton).toBeDefined()
 			expect(screen.queryByText(/%$/)).not.toBeInTheDocument()
 		})
+
+		it("should not display context progress when the context window is infinite", () => {
+			mockModelInfo = { contextWindow: Number.POSITIVE_INFINITY, maxTokens: 200 }
+
+			renderTaskHeader()
+
+			expect(screen.queryByTestId("context-tokens-count")).not.toBeInTheDocument()
+			expect(screen.queryByTestId("context-window-size")).not.toBeInTheDocument()
+			expect(screen.queryByText(/%$/)).not.toBeInTheDocument()
+		})
 	})
 })
diff --git a/webview-ui/src/components/settings/CustomModelInfoSettings.tsx b/webview-ui/src/components/settings/CustomModelInfoSettings.tsx
index 52fb15062f..dc39ef8e69 100644
--- a/webview-ui/src/components/settings/CustomModelInfoSettings.tsx
+++ b/webview-ui/src/components/settings/CustomModelInfoSettings.tsx
@@ -30,21 +30,13 @@ const parsePositiveInteger = (value: string): number | undefined => {
 const getEventValue = (event: ValueChangeEvent): string => {
 	const target = event.target
 
-	if (target && "value" in target && typeof target.value === "string") {
-		return target.value
-	}
-
-	return ""
+	return target && "value" in target && typeof target.value === "string" ? target.value : ""
 }
 
 const getCheckboxValue = (event: ValueChangeEvent): boolean => {
 	const target = event.target
 
-	if (target && "checked" in target && typeof target.checked === "boolean") {
-		return target.checked
-	}
-
-	return false
+	return target && "checked" in target && typeof target.checked === "boolean" ? target.checked : false
 }
 
 const getInputBorderColor = (value: string): string | undefined => {
diff --git a/webview-ui/src/components/settings/__tests__/CustomModelInfoSettings.spec.tsx b/webview-ui/src/components/settings/__tests__/CustomModelInfoSettings.spec.tsx
index 36d9537772..1966bf8860 100644
--- a/webview-ui/src/components/settings/__tests__/CustomModelInfoSettings.spec.tsx
+++ b/webview-ui/src/components/settings/__tests__/CustomModelInfoSettings.spec.tsx
@@ -64,4 +64,61 @@ describe("CustomModelInfoSettings", () => {
 		expect(maxTokensInput).toHaveAttribute("aria-invalid", "true")
 		expect(setApiConfigurationField).toHaveBeenLastCalledWith("customModelInfo", undefined)
 	})
+
+	it("updates capability overrides and warns when output exceeds the context window", () => {
+		const setApiConfigurationField = vi.fn()
+
+		render(
+			<CustomModelInfoSettings
+				apiConfiguration={{
+					apiProvider: "unbound",
+					customModelInfo: { contextWindow: 1000, maxTokens: 2000 },
+				}}
+				setApiConfigurationField={setApiConfigurationField}
+				selectedModelInfo={modelInfo}
+			/>,
+		)
+
+		fireEvent.click(screen.getByText("settings:providers.customModelInfo.title"))
+
+		expect(screen.getByText("settings:providers.customModelInfo.maxTokensWarning")).toBeInTheDocument()
+
+		fireEvent.click(screen.getByText("settings:providers.customModelInfo.supportsImages.label"))
+		expect(setApiConfigurationField).toHaveBeenLastCalledWith("customModelInfo", {
+			contextWindow: 1000,
+			maxTokens: 2000,
+			supportsImages: true,
+		})
+
+		fireEvent.click(screen.getByText("settings:providers.customModelInfo.supportsPromptCache.label"))
+		expect(setApiConfigurationField).toHaveBeenLastCalledWith("customModelInfo", {
+			contextWindow: 1000,
+			maxTokens: 2000,
+			supportsPromptCache: false,
+		})
+	})
+
+	it("syncs externally updated numeric overrides into the inputs", () => {
+		const setApiConfigurationField = vi.fn()
+
+		const { rerender } = render(
+			<CustomModelInfoSettings
+				apiConfiguration={{ apiProvider: "openrouter" }}
+				setApiConfigurationField={setApiConfigurationField}
+				selectedModelInfo={modelInfo}
+			/>,
+		)
+
+		fireEvent.click(screen.getByText("settings:providers.customModelInfo.title"))
+
+		rerender(
+			<CustomModelInfoSettings
+				apiConfiguration={{ apiProvider: "openrouter", customModelInfo: { maxTokens: 200 } }}
+				setApiConfigurationField={setApiConfigurationField}
+				selectedModelInfo={modelInfo}
+			/>,
+		)
+
+		expect(screen.getByLabelText("settings:providers.customModelInfo.maxTokens.label")).toHaveValue("200")
+	})
 })
diff --git a/webview-ui/src/components/ui/hooks/__tests__/useSelectedModel.spec.ts b/webview-ui/src/components/ui/hooks/__tests__/useSelectedModel.spec.ts
index e1af24a3c7..af8accb7cb 100644
--- a/webview-ui/src/components/ui/hooks/__tests__/useSelectedModel.spec.ts
+++ b/webview-ui/src/components/ui/hooks/__tests__/useSelectedModel.spec.ts
@@ -1,13 +1,15 @@
 // npx vitest src/components/ui/hooks/__tests__/useSelectedModel.spec.ts
 
 import React from "react"
-import { QueryClient, QueryClientProvider } from "@tanstack/react-query"
+import { QueryClient, QueryClientProvider, type UseQueryResult } from "@tanstack/react-query"
 import { renderHook } from "@testing-library/react"
 import type { Mock } from "vitest"
 
 import {
 	ProviderSettings,
 	ModelInfo,
+	type RouterModels,
+	type ProviderName,
 	anthropicModels,
 	BEDROCK_1M_CONTEXT_MODEL_IDS,
 	litellmDefaultModelInfo,
@@ -37,6 +39,147 @@ vi.mock("../useOpenRouterModelProviders")
 const mockUseRouterModels = useRouterModels as Mock<typeof useRouterModels>
 const mockUseOpenRouterModelProviders = useOpenRouterModelProviders as Mock<typeof useOpenRouterModelProviders>
 
+type OpenRouterModelProviders = NonNullable<ReturnType<typeof useOpenRouterModelProviders>["data"]>
+
+const emptyRouterModels: RouterModels = {
+	openrouter: {},
+	"vercel-ai-gateway": {},
+	"zoo-gateway": {},
+	litellm: {},
+	requesty: {},
+	unbound: {},
+	poe: {},
+	deepseek: {},
+	moonshot: {},
+	"opencode-go": {},
+	kenari: {},
+	"kimi-code": {},
+	ollama: {},
+	lmstudio: {},
+}
+
+const routerProviderCases = [
+	{
+		provider: providerIdentifiers.openrouter,
+		modelKey: "openrouter",
+		modelId: "openrouter/future-model",
+		settings: {
+			apiProvider: providerIdentifiers.openrouter,
+			openRouterModelId: "openrouter/future-model",
+		},
+	},
+	{
+		provider: providerIdentifiers.requesty,
+		modelKey: "requesty",
+		modelId: "requesty/future-model",
+		settings: {
+			apiProvider: providerIdentifiers.requesty,
+			requestyModelId: "requesty/future-model",
+		},
+	},
+	{
+		provider: providerIdentifiers.unbound,
+		modelKey: "unbound",
+		modelId: "unbound/future-model",
+		settings: {
+			apiProvider: providerIdentifiers.unbound,
+			unboundModelId: "unbound/future-model",
+		},
+	},
+	{
+		provider: providerIdentifiers.vercelAiGateway,
+		modelKey: "vercel-ai-gateway",
+		modelId: "vercel/future-model",
+		settings: {
+			apiProvider: providerIdentifiers.vercelAiGateway,
+			vercelAiGatewayModelId: "vercel/future-model",
+		},
+	},
+	{
+		provider: providerIdentifiers.zooGateway,
+		modelKey: "zoo-gateway",
+		modelId: "zoo/future-model",
+		settings: {
+			apiProvider: providerIdentifiers.zooGateway,
+			zooGatewayModelId: "zoo/future-model",
+		},
+	},
+] as const satisfies ReadonlyArray<{
+	provider: ProviderName
+	modelKey: keyof RouterModels
+	modelId: string
+	settings: ProviderSettings
+}>
+
+const createRouterModels = (modelKey: keyof RouterModels, modelId: string, info?: ModelInfo): RouterModels => {
+	const models = { ...emptyRouterModels }
+	models[modelKey] = info ? { [modelId]: info } : {}
+	return models
+}
+
+const createQueryResult = <TData>(
+	data: TData | undefined,
+	fallbackData: TData,
+	isLoading: boolean,
+): UseQueryResult<TData, Error> =>
+	isLoading
+		? {
+				data: undefined,
+				dataUpdatedAt: 0,
+				error: null,
+				errorUpdatedAt: 0,
+				failureCount: 0,
+				failureReason: null,
+				errorUpdateCount: 0,
+				isError: false,
+				isFetched: false,
+				isFetchedAfterMount: false,
+				isFetching: true,
+				isLoading: true,
+				isPending: true,
+				isLoadingError: false,
+				isInitialLoading: true,
+				isPaused: false,
+				isPlaceholderData: false,
+				isRefetchError: false,
+				isRefetching: false,
+				isStale: false,
+				isSuccess: false,
+				isEnabled: true,
+				refetch: vi.fn(),
+				status: "pending",
+				fetchStatus: "fetching",
+				promise: Promise.resolve(fallbackData),
+			}
+		: {
+				data: data ?? fallbackData,
+				dataUpdatedAt: 0,
+				error: null,
+				errorUpdatedAt: 0,
+				failureCount: 0,
+				failureReason: null,
+				errorUpdateCount: 0,
+				isError: false,
+				isFetched: true,
+				isFetchedAfterMount: true,
+				isFetching: false,
+				isLoading: false,
+				isPending: false,
+				isLoadingError: false,
+				isInitialLoading: false,
+				isPaused: false,
+				isPlaceholderData: false,
+				isRefetchError: false,
+				isRefetching: false,
+				isStale: false,
+				isSuccess: true,
+				isEnabled: true,
+				refetch: vi.fn(),
+				status: "success",
+				fetchStatus: "idle",
+				promise: Promise.resolve(data ?? fallbackData),
+			}
+
 const createWrapper = () => {
 	const queryClient = new QueryClient({
 		defaultOptions: {
@@ -307,18 +450,80 @@ describe("useSelectedModel", () => {
 	})
 
 	describe("loading and error states", () => {
+		it.each(routerProviderCases)(
+			"preserves the configured %s model ID and applies custom metadata while data is loading",
+			({ provider, modelId, settings }) => {
+				mockUseRouterModels.mockReturnValue(createQueryResult(undefined, emptyRouterModels, true))
+				mockUseOpenRouterModelProviders.mockReturnValue(
+					createQueryResult<OpenRouterModelProviders>({}, {}, false),
+				)
+
+				const apiConfiguration: ProviderSettings = {
+					...settings,
+					customModelInfo: {
+						contextWindow: 100_000,
+						maxTokens: 10_000,
+						supportsImages: true,
+					},
+				}
+
+				const wrapper = createWrapper()
+				const { result } = renderHook(() => useSelectedModel(apiConfiguration), { wrapper })
+
+				expect(result.current.provider).toBe(provider)
+				expect(result.current.id).toBe(modelId)
+				expect(result.current.info).toMatchObject({
+					contextWindow: 100_000,
+					maxTokens: 10_000,
+					supportsImages: true,
+				})
+			},
+		)
+
+		it.each(routerProviderCases)(
+			"applies custom metadata to a listed %s model",
+			({ provider, modelKey, modelId, settings }) => {
+				const discoveredInfo: ModelInfo = {
+					contextWindow: 8192,
+					maxTokens: 4096,
+					supportsImages: false,
+					supportsPromptCache: false,
+				}
+
+				mockUseRouterModels.mockReturnValue(
+					createQueryResult(createRouterModels(modelKey, modelId, discoveredInfo), emptyRouterModels, false),
+				)
+				mockUseOpenRouterModelProviders.mockReturnValue(
+					createQueryResult<OpenRouterModelProviders>({}, {}, false),
+				)
+
+				const apiConfiguration: ProviderSettings = {
+					...settings,
+					customModelInfo: {
+						contextWindow: 100_000,
+						maxTokens: 10_000,
+						supportsImages: true,
+					},
+				}
+
+				const wrapper = createWrapper()
+				const { result } = renderHook(() => useSelectedModel(apiConfiguration), { wrapper })
+
+				expect(result.current.provider).toBe(provider)
+				expect(result.current.id).toBe(modelId)
+				expect(result.current.info).toMatchObject({
+					contextWindow: 100_000,
+					maxTokens: 10_000,
+					supportsImages: true,
+					supportsPromptCache: false,
+				})
+			},
+		)
+
 		it("preserves a router model ID and applies custom metadata while model data is loading", () => {
-			mockUseRouterModels.mockReturnValue({
-				data: undefined,
-				isLoading: true,
-				isError: false,
-			} as any)
+			mockUseRouterModels.mockReturnValue(createQueryResult(undefined, emptyRouterModels, true))
 
-			mockUseOpenRouterModelProviders.mockReturnValue({
-				data: undefined,
-				isLoading: false,
-				isError: false,
-			} as any)
+			mockUseOpenRouterModelProviders.mockReturnValue(createQueryResult<OpenRouterModelProviders>({}, {}, false))
 
 			const apiConfiguration: ProviderSettings = {
 				apiProvider: providerIdentifiers.openrouter,
diff --git a/webview-ui/src/i18n/locales/ca/settings.json b/webview-ui/src/i18n/locales/ca/settings.json
index 7c0454f55c..bf8f256b96 100644
--- a/webview-ui/src/i18n/locales/ca/settings.json
+++ b/webview-ui/src/i18n/locales/ca/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Restablir als valors per defecte"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Límit de freqüència",
 			"description": "Temps mínim entre sol·licituds d'API."
diff --git a/webview-ui/src/i18n/locales/de/settings.json b/webview-ui/src/i18n/locales/de/settings.json
index 504bb56cba..bc90b462f0 100644
--- a/webview-ui/src/i18n/locales/de/settings.json
+++ b/webview-ui/src/i18n/locales/de/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Auf Standardwerte zurücksetzen"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Ratenbegrenzung",
 			"description": "Minimale Zeit zwischen API-Anfragen."
diff --git a/webview-ui/src/i18n/locales/es/settings.json b/webview-ui/src/i18n/locales/es/settings.json
index eba338005f..f365cafe83 100644
--- a/webview-ui/src/i18n/locales/es/settings.json
+++ b/webview-ui/src/i18n/locales/es/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Restablecer valores predeterminados"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Límite de tasa",
 			"description": "Tiempo mínimo entre solicitudes de API."
diff --git a/webview-ui/src/i18n/locales/fr/settings.json b/webview-ui/src/i18n/locales/fr/settings.json
index d6e6e0e64e..aa4cd1cbb2 100644
--- a/webview-ui/src/i18n/locales/fr/settings.json
+++ b/webview-ui/src/i18n/locales/fr/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Réinitialiser les valeurs par défaut"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Limite de débit",
 			"description": "Temps minimum entre les requêtes API."
diff --git a/webview-ui/src/i18n/locales/hi/settings.json b/webview-ui/src/i18n/locales/hi/settings.json
index 3ff02125c5..a4eb5774c9 100644
--- a/webview-ui/src/i18n/locales/hi/settings.json
+++ b/webview-ui/src/i18n/locales/hi/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "डिफ़ॉल्ट पर रीसेट करें"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "दर सीमा",
 			"description": "API अनुरोधों के बीच न्यूनतम समय।"
diff --git a/webview-ui/src/i18n/locales/id/settings.json b/webview-ui/src/i18n/locales/id/settings.json
index 6c4b91243f..81e2d24b04 100644
--- a/webview-ui/src/i18n/locales/id/settings.json
+++ b/webview-ui/src/i18n/locales/id/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Reset ke Default"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Rate limit",
 			"description": "Waktu minimum antara permintaan API."
diff --git a/webview-ui/src/i18n/locales/it/settings.json b/webview-ui/src/i18n/locales/it/settings.json
index 8f7fd7e917..49aaac46bc 100644
--- a/webview-ui/src/i18n/locales/it/settings.json
+++ b/webview-ui/src/i18n/locales/it/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Ripristina valori predefiniti"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Limite di frequenza",
 			"description": "Tempo minimo tra le richieste API."
diff --git a/webview-ui/src/i18n/locales/ja/settings.json b/webview-ui/src/i18n/locales/ja/settings.json
index ab692a49f8..119783f39e 100644
--- a/webview-ui/src/i18n/locales/ja/settings.json
+++ b/webview-ui/src/i18n/locales/ja/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "デフォルトにリセット"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "レート制限",
 			"description": "APIリクエスト間の最小時間。"
diff --git a/webview-ui/src/i18n/locales/ko/settings.json b/webview-ui/src/i18n/locales/ko/settings.json
index 4e44f8170d..731ba5fdb8 100644
--- a/webview-ui/src/i18n/locales/ko/settings.json
+++ b/webview-ui/src/i18n/locales/ko/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "기본값으로 재설정"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "속도 제한",
 			"description": "API 요청 간 최소 시간."
diff --git a/webview-ui/src/i18n/locales/nl/settings.json b/webview-ui/src/i18n/locales/nl/settings.json
index d517df4bd0..d49b5fe29c 100644
--- a/webview-ui/src/i18n/locales/nl/settings.json
+++ b/webview-ui/src/i18n/locales/nl/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Standaardwaarden herstellen"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Snelheidslimiet",
 			"description": "Minimale tijd tussen API-verzoeken."
diff --git a/webview-ui/src/i18n/locales/pl/settings.json b/webview-ui/src/i18n/locales/pl/settings.json
index 3ef8e06c32..b753b82c9a 100644
--- a/webview-ui/src/i18n/locales/pl/settings.json
+++ b/webview-ui/src/i18n/locales/pl/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Przywróć domyślne"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Limit szybkości",
 			"description": "Minimalny czas między żądaniami API."
diff --git a/webview-ui/src/i18n/locales/pt-BR/settings.json b/webview-ui/src/i18n/locales/pt-BR/settings.json
index 9c67418d16..dfdadefb92 100644
--- a/webview-ui/src/i18n/locales/pt-BR/settings.json
+++ b/webview-ui/src/i18n/locales/pt-BR/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Restaurar Padrões"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Limite de taxa",
 			"description": "Tempo mínimo entre requisições de API."
diff --git a/webview-ui/src/i18n/locales/ru/settings.json b/webview-ui/src/i18n/locales/ru/settings.json
index 6d81073dbe..234db259eb 100644
--- a/webview-ui/src/i18n/locales/ru/settings.json
+++ b/webview-ui/src/i18n/locales/ru/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Сбросить к значениям по умолчанию"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Лимит скорости",
 			"description": "Минимальное время между запросами к API."
diff --git a/webview-ui/src/i18n/locales/tr/settings.json b/webview-ui/src/i18n/locales/tr/settings.json
index 0456367efc..bf3504afdb 100644
--- a/webview-ui/src/i18n/locales/tr/settings.json
+++ b/webview-ui/src/i18n/locales/tr/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Varsayılanlara Sıfırla"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Hız sınırı",
 			"description": "API istekleri arasındaki minimum süre."
diff --git a/webview-ui/src/i18n/locales/vi/settings.json b/webview-ui/src/i18n/locales/vi/settings.json
index 4beb3f7171..4a247b12f6 100644
--- a/webview-ui/src/i18n/locales/vi/settings.json
+++ b/webview-ui/src/i18n/locales/vi/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "Đặt lại về mặc định"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "Giới hạn tốc độ",
 			"description": "Thời gian tối thiểu giữa các yêu cầu API."
diff --git a/webview-ui/src/i18n/locales/zh-CN/settings.json b/webview-ui/src/i18n/locales/zh-CN/settings.json
index 8624c1899b..8990783fa5 100644
--- a/webview-ui/src/i18n/locales/zh-CN/settings.json
+++ b/webview-ui/src/i18n/locales/zh-CN/settings.json
@@ -610,6 +610,29 @@
 			},
 			"resetDefaults": "重置为默认值"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "API 请求频率限制",
 			"description": "设置API请求的最小间隔时间"
diff --git a/webview-ui/src/i18n/locales/zh-TW/settings.json b/webview-ui/src/i18n/locales/zh-TW/settings.json
index 8556e8b2f4..290b427107 100644
--- a/webview-ui/src/i18n/locales/zh-TW/settings.json
+++ b/webview-ui/src/i18n/locales/zh-TW/settings.json
@@ -637,6 +637,29 @@
 			},
 			"resetDefaults": "重設為預設值"
 		},
+		"customModelInfo": {
+			"title": "Custom model metadata",
+			"description": "Override context and capability metadata when the provider cannot detect your model accurately.",
+			"unresolved": "Model metadata is unavailable. Enter the context window to enable accurate token tracking.",
+			"contextWindow": {
+				"label": "Context window",
+				"description": "Total tokens the model can process, including input and output."
+			},
+			"maxTokens": {
+				"label": "Max output tokens",
+				"description": "Maximum number of tokens the model can generate in one response."
+			},
+			"supportsImages": {
+				"label": "Supports images",
+				"description": "Override whether the model accepts image content."
+			},
+			"supportsPromptCache": {
+				"label": "Supports prompt caching",
+				"description": "Override whether prompt caching is supported."
+			},
+			"maxTokensWarning": "Max output tokens exceed the context window.",
+			"reset": "Reset to detected values"
+		},
 		"rateLimitSeconds": {
 			"label": "速率限制",
 			"description": "API 請求間的最短時間"