feat: Qwen3.8 Max GA — switch to GA model id, restore pay-go default

The GA release drops the -preview suffix and opens pay-as-you-go access.
Verified live before the change: the token-plan endpoint answers 200 for
qwen3.8-max (old qwen3.8-max-preview still answers but is deprecated),
and the pay-go dashscope Anthropic endpoint now returns 200 with a
pay-go key — the exact condition v1.11.1 said would restore the pay-go
variant. So qwen gets a 3.8-max variant back as its default (bare
aliases qwen/tongyi and qwen3.8/qwen3.8-max point at it, qwen-max/
qwen3.7 stay on 3.7-max), and qwen-plan-3.8-max switches to the GA id.

Post-change doctor through the user's runtime config: qwen3.8-max [OK]
1.47s, qwen3.7-max [OK] 4.81s, qwen-plan-3.8-max [OK] 1.01s.

Author: 鲁工

README.md

@@ -197,8 +197,9 @@ ccmr claude --print --output-format json "你的问题"
 | `kimi-plan-for-coding-highspeed` | `kimi-plan-highspeed` | Kimi K2.7 Code HighSpeed（Allegretto 及以上） | Kimi Code 会员订阅 |
 | `minimax-m3` | `minimax`, `minimax-cn`, `mm` | MiniMax M3 | MiniMax CN |
 | `minimax-global-m3` | `minimax-global`, `minimax-io` | MiniMax M3 | MiniMax Global |
-| `qwen3.7-max` | `qwen`, `qwen-max`, `qwen3.7`, `tongyi` | Qwen3.7 Max | 阿里云（按量付费） |
-| `qwen-plan-3.8-max` | `qwen-plan`, `qwen-plan-3.8`, `qwen-plan-max`, `qwen3.8`, `qwen3.8-max` | Qwen3.8 Max (Preview) | 千问 Token Plan（订阅） |
+| `qwen3.8-max` | `qwen`, `tongyi`, `qwen3.8` | Qwen3.8 Max | 阿里云（按量付费） |
+| `qwen3.7-max` | `qwen-max`, `qwen3.7` | Qwen3.7 Max | 阿里云（按量付费） |
+| `qwen-plan-3.8-max` | `qwen-plan`, `qwen-plan-3.8`, `qwen-plan-max` | Qwen3.8 Max | 千问 Token Plan（订阅） |
 | `qwen-plan-3.7-max` | `qwen-plan-3.7` | Qwen3.7 Max | 千问 Token Plan（订阅） |
 | `glm-plan-5.2` | `glm-plan`, `glm`, `glm-5.2`, `zhipu`, `chatglm` | GLM-5.2 | 智谱 GLM Coding Plan（订阅） |
 | `glm-global-5.2` | `glm-global`, `zai`, `z-ai` | GLM-5.2 | Z.ai（国际） |
@@ -226,7 +227,7 @@ ccmr claude --print --output-format json "你的问题"
 | Kimi K3 (Coding Plan `kimi-plan-k3-1m` / `kimi-plan-k3`) | 1M / 256K（按档位） | 1M / 128K |
 | Kimi K2.7 Code / HighSpeed (Coding Plan) | 256K | 32K |
 | MiniMax M3 (CN / Global) | 1M | 128K |
-| Qwen3.8 Max Preview (仅 Token Plan) | 1M | 64K |
+| Qwen3.8 Max (按量付费 / Token Plan) | 1M | 64K |
 | Qwen3.7 Max (按量付费 / Token Plan) | 1M | 64K |
 | GLM-5.2 (Coding Plan / 国际) | 1M | 128K |
 | Step 3.7 Flash (按量付费 / Step Plan) | 256K | 384K |
@@ -405,8 +406,8 @@ npx claude-code-model-router claude
 ```bash
 # 使用短名称（向后兼容）
 /model deepseek   # 切换到 DeepSeek V4 Pro
-/model qwen       # 切换到 Qwen3.7 Max（按量付费）
-/model qwen-plan  # 切换到 Qwen3.8 Max Preview（千问 Token Plan 订阅）
+/model qwen       # 切换到 Qwen3.8 Max（按量付费）
+/model qwen-plan  # 切换到 Qwen3.8 Max（千问 Token Plan 订阅）
 /model glm        # 切换到 GLM-5.2（智谱 Coding Plan 订阅）
 /model glm-global # 切换到 GLM-5.2（国际 Z.ai）
 /model seed       # 切换到 Doubao Seed 2.1 Pro（火山方舟 按量付费）
@@ -434,8 +435,9 @@ npx claude-code-model-router claude
 /model kimi-k2.6                 # Kimi K2.6
 /model kimi-k2.7-code            # Kimi K2.7 Code
 /model kimi-k2.7-code-highspeed  # Kimi K2.7 Code HighSpeed
+/model qwen3.8-max               # Qwen3.8 Max（按量付费；别名 qwen / qwen3.8）
 /model qwen3.7-max               # Qwen3.7 Max（按量付费）
-/model qwen-plan-3.8-max         # Qwen3.8 Max Preview（Token Plan 订阅；别名 qwen3.8 / qwen3.8-max）
+/model qwen-plan-3.8-max         # Qwen3.8 Max（Token Plan 订阅）
 /model qwen-plan-3.7-max         # Qwen3.7 Max（Token Plan 订阅）
 /model seed-2.1-pro              # Doubao Seed 2.1 Pro（按量付费）
 /model seed-2.1-turbo            # Doubao Seed 2.1 Turbo（按量付费）
@@ -524,6 +526,11 @@ DeepSeek Anthropic 兼容接口会忽略 `metadata` 字段，但某些 Claude Co
 
 ## 更新日志
 
+### v1.14.0
+
+- **Qwen3.8 Max 正式版（GA）**：模型 ID 去掉 `-preview` 后缀，统一更新为 `qwen3.8-max`（旧 `qwen3.8-max-preview` 上游暂可用但已弃用），`qwen-plan-3.8-max` 同步切到 GA id
+- **恢复按量付费 `qwen3.8-max` 并设为默认**：正式版已开放按量付费（实测按量 key 在 dashscope Anthropic 端点返回 200，兑现 v1.11.1 的"待开放后恢复"）。裸别名 `qwen` / `tongyi` 及 `qwen3.8` / `qwen3.8-max` 现指向按量版 3.8；`qwen3.7-max`（别名 `qwen-max` / `qwen3.7`）保留
+
 ### v1.13.0
 
 - 移除 **GLM-5.1**（国内 `glm-plan-5.1` 与国际 `glm-global-5.1`，别名 `glm-5` / `glm-5.1` 同步下线）：智谱官方已退役 5.1，Coding Plan 对旧模型的调用会被上游自动切换到 GLM-5.2，继续保留只会误导。GLM 现仅保留 5.2（国内 Coding Plan / 国际 Z.ai）

dist/config.d.ts.map

@@ -1,6 +1,6 @@
 {
   "name": "claude-code-model-router",
-  "version": "1.13.0",
+  "version": "1.14.0",
   "description": "A lightweight API gateway for routing Claude Code requests to multiple AI models",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",

dist/config.js

@@ -203,12 +203,18 @@ export const DEFAULT_CONFIG: RouterConfig = {
       auth_type: 'api_key',
       supports_streaming: true,
       supports_tools: true,
-      default_variant: '3.7-max',
+      default_variant: '3.8-max',
       variants: {
-        // qwen3.8-max-preview is Token-Plan-only today; a pay-as-you-go key
-        // gets 403 Model.AccessDenied. It lives on the qwen-plan provider
-        // (sk-sp- subscription key) below, not here. Re-add a pay-go variant
-        // only once Alibaba opens the preview to pay-as-you-go.
+        // GA release of Qwen3.8 Max dropped the -preview suffix and opened
+        // pay-as-you-go access (the preview was Token-Plan-only and 403'd
+        // pay-go keys; verified live that the GA id returns 200 here).
+        '3.8-max': {
+          model_key: 'qwen3.8-max',
+          display_name: 'Qwen3.8 Max',
+          model_id: 'qwen3.8-max',
+          max_tokens: 65536,
+          context_window: 1000000,
+        },
         '3.7-max': {
           model_key: 'qwen3.7-max',
           display_name: 'Qwen3.7 Max',
@@ -234,7 +240,9 @@ export const DEFAULT_CONFIG: RouterConfig = {
       variants: {
         '3.8-max': {
           display_name: 'Qwen3.8 Max (Token Plan)',
-          model_id: 'qwen3.8-max-preview',
+          // GA id; the retired qwen3.8-max-preview id still answers upstream
+          // but is deprecated.
+          model_id: 'qwen3.8-max',
           max_tokens: 65536,
           context_window: 1000000,
         },
@@ -525,14 +533,13 @@ export const DEFAULT_CONFIG: RouterConfig = {
     'minimax-io': 'minimax-global-m3',
     'minimax-global-m3': 'minimax-global-m3',
     mm: 'minimax-m3',
-    qwen: 'qwen3.7-max',
-    tongyi: 'qwen3.7-max',
+    qwen: 'qwen3.8-max',
+    tongyi: 'qwen3.8-max',
+    'qwen3.8': 'qwen3.8-max',
+    'qwen3.8-max': 'qwen3.8-max',
     'qwen-max': 'qwen3.7-max',
     'qwen3.7-max': 'qwen3.7-max',
     'qwen3.7': 'qwen3.7-max',
-    // qwen3.8-max-preview is reachable only via the Token Plan provider below.
-    'qwen3.8': 'qwen-plan-3.8-max',
-    'qwen3.8-max': 'qwen-plan-3.8-max',
     'qwen-plan': 'qwen-plan-3.8-max',
     'qwen-plan-3.8': 'qwen-plan-3.8-max',
     'qwen-plan-3.8-max': 'qwen-plan-3.8-max',
@@ -1287,9 +1294,15 @@ providers:
     api_key_env: QWEN_API_KEY
     auth_header: x-api-key
     auth_type: api_key
-    default_variant: 3.7-max
+    default_variant: 3.8-max
     variants:
-      # qwen3.8-max-preview 目前仅 Token Plan 发放，按量付费调用 403，见下方 qwen-plan
+      # Qwen3.8 Max 正式版（GA 去掉 -preview 后缀）已开放按量付费，实测 200
+      3.8-max:
+        model_key: qwen3.8-max
+        display_name: "Qwen3.8 Max"
+        model_id: qwen3.8-max
+        max_tokens: 65536
+        context_window: 1000000
       3.7-max:
         model_key: qwen3.7-max
         display_name: "Qwen3.7 Max"
@@ -1310,7 +1323,8 @@ providers:
     variants:
       3.8-max:
         display_name: "Qwen3.8 Max (Token Plan)"
-        model_id: qwen3.8-max-preview
+        # GA id；旧 qwen3.8-max-preview 上游暂可用但已弃用
+        model_id: qwen3.8-max
         max_tokens: 65536
         context_window: 1000000
       3.7-max:
@@ -1548,13 +1562,13 @@ aliases:
   minimax-io: minimax-global-m3
   minimax-global-m3: minimax-global-m3
   mm: minimax-m3
-  qwen: qwen3.7-max
-  tongyi: qwen3.7-max
+  qwen: qwen3.8-max
+  tongyi: qwen3.8-max
+  qwen3.8: qwen3.8-max
+  qwen3.8-max: qwen3.8-max
   qwen-max: qwen3.7-max
   qwen3.7-max: qwen3.7-max
   qwen3.7: qwen3.7-max
-  qwen3.8: qwen-plan-3.8-max
-  qwen3.8-max: qwen-plan-3.8-max
   qwen-plan: qwen-plan-3.8-max
   qwen-plan-3.8: qwen-plan-3.8-max
   qwen-plan-3.8-max: qwen-plan-3.8-max

dist/config.js.map

@@ -154,21 +154,27 @@ describe('Kimi Code: coding-plan subscription provider', () => {
   });
 });
 
-describe('Qwen 3.8: Token-Plan-only, 3.5 removal, pay-go stays on 3.7', () => {
+describe('Qwen 3.8 GA: pay-go restored as default, Token Plan on the GA id', () => {
   const manager = new ConfigManager(null);
   const config = manager.getConfig();
 
-  it('keeps the pay-as-you-go qwen provider on 3.7-max (3.8 is not pay-go)', () => {
-    // qwen3.8-max-preview returns 403 on pay-as-you-go keys (Token-Plan-only),
-    // so it must NOT be a pay-go model; the bare qwen alias stays on 3.7.
-    expect(config.models['qwen3.8-max']).toBeUndefined();
-    const model = config.models['qwen3.7-max'];
+  it('restores qwen3.8-max as the pay-as-you-go default', () => {
+    // The GA release dropped the -preview suffix AND opened pay-as-you-go
+    // access (verified live: dashscope endpoint returns 200 with a pay-go
+    // key), so the v1.11.1 removal is reverted: 3.8 is back as the pay-go
+    // default and the bare aliases point at it.
+    const model = config.models['qwen3.8-max'];
     expect(model).toBeDefined();
-    expect(model.model_id).toBe('qwen3.7-max');
+    expect(model.model_id).toBe('qwen3.8-max');
     expect(model.api_key_env).toBe('QWEN_API_KEY');
-    expect(manager.resolveModelName('qwen')).toBe('qwen3.7-max');
-    expect(manager.resolveModelName('tongyi')).toBe('qwen3.7-max');
+    expect(manager.resolveModelName('qwen')).toBe('qwen3.8-max');
+    expect(manager.resolveModelName('tongyi')).toBe('qwen3.8-max');
+    expect(manager.resolveModelName('qwen3.8')).toBe('qwen3.8-max');
+    expect(manager.resolveModelName('qwen3.8-max')).toBe('qwen3.8-max');
+    // 3.7 stays available under its versioned aliases
+    expect(config.models['qwen3.7-max'].model_id).toBe('qwen3.7-max');
     expect(manager.resolveModelName('qwen-max')).toBe('qwen3.7-max');
+    expect(manager.resolveModelName('qwen3.7')).toBe('qwen3.7-max');
   });
 
   it('removes the qwen3.5 models and their alias', () => {
@@ -178,23 +184,20 @@ describe('Qwen 3.8: Token-Plan-only, 3.5 removal, pay-go stays on 3.7', () => {
     expect(config.models[manager.resolveModelName('qwen3.5')]).toBeUndefined();
   });
 
-  it('serves qwen3.8-max-preview only through the Token Plan provider', () => {
+  it('keeps the Token Plan provider on its dedicated endpoint with the GA id', () => {
     // sk-sp- keys from platform.qianwenai.com only work on the dedicated
-    // token-plan endpoint (they 403 on the pay-as-you-go dashscope one), so
-    // the subscription is its own provider/key + base_url — and the only
-    // place the preview model actually resolves.
+    // token-plan endpoint (they 403 on the pay-as-you-go dashscope one).
+    // The subscription model now uses the GA id too (the old -preview id
+    // still answers upstream but is deprecated).
     const model = config.models['qwen-plan-3.8-max'];
     expect(model).toBeDefined();
-    expect(model.model_id).toBe('qwen3.8-max-preview');
+    expect(model.model_id).toBe('qwen3.8-max');
     expect(model.base_url).toBe(
       'https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic'
     );
     expect(model.api_key_env).toBe('QWEN_PLAN_API_KEY');
     expect(model.auth_type).toBe('api_key');
     expect(manager.resolveModelName('qwen-plan')).toBe('qwen-plan-3.8-max');
-    // the intuitive 3.8 names route to the subscription model (its only home)
-    expect(manager.resolveModelName('qwen3.8')).toBe('qwen-plan-3.8-max');
-    expect(manager.resolveModelName('qwen3.8-max')).toBe('qwen-plan-3.8-max');
     expect(config.models['qwen-plan-3.7-max'].model_id).toBe('qwen3.7-max');
     expect(config.models['qwen-plan-3.7-max'].api_key_env).toBe('QWEN_PLAN_API_KEY');
   });