feat(kosong): honor explicit anthropic max output override

.changeset/honor-explicit-anthropic-max-output.md

@@ -0,0 +1,5 @@
+---
+"@moonshot-ai/kimi-code": patch
+---
+
+Honor explicit Anthropic `max_output_size` settings instead of clamping them to built-in ceilings.

docs/en/configuration/config-files.md

@@ -125,7 +125,7 @@ Each entry in the `models` table defines a model alias (the name used in `defaul
 | `provider` | `string` | Yes | Name of the provider to use; must be defined in `providers` |
 | `model` | `string` | Yes | Model identifier sent to the server when calling the API |
 | `max_context_size` | `integer` | Yes | Maximum context length in tokens; must be at least 1 |
-| `max_output_size` | `integer` | No | Per-request output token cap (maps to `max_tokens`). Currently only the `anthropic` provider honors it; recognized Claude models are automatically clamped to the server-side maximum |
+| `max_output_size` | `integer` | No | Per-request output token cap (maps to `max_tokens`). Currently only the `anthropic` provider honors it. When set for a Claude model, this explicit value overrides the built-in server-side maximum |
 | `capabilities` | `array<string>` | No | Capability tags to add explicitly: `thinking`, `image_in`, `video_in`, `audio_in`, `tool_use`. Unioned with the capabilities auto-detected by the provider — entries can only be added, never removed |
 | `support_efforts` | `array<string>` | No | Thinking effort levels declared by the model catalog. Managed and open-platform refreshes may rewrite this field; to pin it manually, set `[models."<alias>".overrides] support_efforts` instead |
 | `default_effort` | `string` | No | Default thinking effort for the model. Managed and open-platform refreshes may rewrite this field; to pin it manually, set `[models."<alias>".overrides] default_effort` instead |

docs/en/configuration/env-vars.md

@@ -107,7 +107,7 @@ Complete variable list:
 | `KIMI_MODEL_MAX_CONTEXT_SIZE` | No | Maximum context length (tokens) | `262144` (256 K) |
 | `KIMI_MODEL_CAPABILITIES` | No | Comma-separated capability tags, unioned with auto-detected capabilities | `image_in,thinking` |
 | `KIMI_MODEL_DISPLAY_NAME` | No | Name shown in `/model` | Falls back to `KIMI_MODEL_NAME` |
-| `KIMI_MODEL_MAX_OUTPUT_SIZE` | No | Per-request output cap (`anthropic` only) | Model default |
+| `KIMI_MODEL_MAX_OUTPUT_SIZE` | No | Per-request output cap (`anthropic` only); when set, overrides the built-in Claude ceiling | Model default |
 | `KIMI_MODEL_REASONING_KEY` | No | Reasoning field name override (`openai` only) | Auto-detected |
 | `KIMI_MODEL_THINKING_EFFORT` | No | Thinking effort level: `low`/`medium`/`high`/`xhigh`/`max` | — |
 | `KIMI_MODEL_ADAPTIVE_THINKING` | No | Force adaptive thinking on or off (`anthropic` only) | Inferred from model name |

docs/zh/configuration/config-files.md

@@ -125,7 +125,7 @@ KIMI_BASE_URL = "https://api.moonshot.ai/v1"
 | `provider` | `string` | 是 | 使用的供应商名称，必须在 `providers` 中定义 |
 | `model` | `string` | 是 | 调用 API 时实际传给服务端的模型 ID |
 | `max_context_size` | `integer` | 是 | 最大上下文长度（token 数），必须 ≥ 1 |
-| `max_output_size` | `integer` | 否 | 单次请求的输出 token 上限（对应 `max_tokens`）。目前仅 `anthropic` 供应商读取；已识别的 Claude 系列会自动限制在服务端允许的最大值内 |
+| `max_output_size` | `integer` | 否 | 单次请求的输出 token 上限（对应 `max_tokens`）。目前仅 `anthropic` 供应商读取。为 Claude 模型设置后，这个显式值会覆盖内置的服务端最大值 |
 | `capabilities` | `array<string>` | 否 | 显式追加的能力标签：`thinking`、`image_in`、`video_in`、`audio_in`、`tool_use`。与供应商自动识别的能力取并集，只能追加不能移除 |
 | `support_efforts` | `array<string>` | 否 | 模型目录声明的 Thinking 档位。managed 和 open-platform 刷新可能会改写该字段；如需手动固定，请改用 `[models."<alias>".overrides] support_efforts` |
 | `default_effort` | `string` | 否 | 模型的默认 Thinking 档位。managed 和 open-platform 刷新可能会改写该字段；如需手动固定，请改用 `[models."<alias>".overrides] default_effort` |

docs/zh/configuration/env-vars.md

@@ -107,7 +107,7 @@ kimi
 | `KIMI_MODEL_MAX_CONTEXT_SIZE` | 否 | 最大上下文长度（token 数） | `262144`（256K） |
 | `KIMI_MODEL_CAPABILITIES` | 否 | 逗号分隔的能力标签，与自动探测的能力取并集 | `image_in,thinking` |
 | `KIMI_MODEL_DISPLAY_NAME` | 否 | 在 `/model` 中显示的名称 | 回退到 `KIMI_MODEL_NAME` |
-| `KIMI_MODEL_MAX_OUTPUT_SIZE` | 否 | 单次输出上限（仅 `anthropic`） | 模型默认值 |
+| `KIMI_MODEL_MAX_OUTPUT_SIZE` | 否 | 单次输出上限（仅 `anthropic`）；设置后会覆盖内置的 Claude 上限 | 模型默认值 |
 | `KIMI_MODEL_REASONING_KEY` | 否 | 推理字段名覆盖（仅 `openai`） | 自动探测 |
 | `KIMI_MODEL_THINKING_EFFORT` | 否 | Thinking 强度：`low`/`medium`/`high`/`xhigh`/`max` | — |
 | `KIMI_MODEL_ADAPTIVE_THINKING` | 否 | 强制开启或关闭 adaptive thinking（仅 `anthropic`） | 按模型名推断 |

packages/kosong/src/providers/anthropic.ts

@@ -985,7 +985,7 @@ export class AnthropicChatProvider implements ChatProvider {
     this._client = this._apiKey === undefined ? undefined : this._buildClient(this._apiKey);
     this._explicitMaxTokens = options.defaultMaxTokens !== undefined;
     this._generationKwargs = {
-      max_tokens: resolveDefaultMaxTokens(options.model, options.defaultMaxTokens),
+      max_tokens: options.defaultMaxTokens ?? resolveDefaultMaxTokens(options.model),
       betaFeatures: options.betaFeatures ?? [INTERLEAVED_THINKING_BETA],
     };
   }

packages/kosong/test/anthropic.test.ts

@@ -2759,8 +2759,8 @@ describe('AnthropicChatProvider constructor max_tokens', () => {
     expect(await maxTokensFor('claude-opus-4-7', { defaultMaxTokens: 200 })).toBe(200);
   });
 
-  it('clamps defaultMaxTokens above the documented ceiling for known models', async () => {
-    expect(await maxTokensFor('claude-opus-4-7', { defaultMaxTokens: 999999 })).toBe(128000);
+  it('honors explicit defaultMaxTokens above the ceiling for known models', async () => {
+    expect(await maxTokensFor('claude-opus-4-7', { defaultMaxTokens: 999999 })).toBe(999999);
   });
 
   it('withMaxCompletionTokens sets max_tokens when no existing cap is present', async () => {
@@ -2847,6 +2847,21 @@ describe('AnthropicChatProvider constructor max_tokens', () => {
     expect(body['max_tokens']).toBe(128000);
   });
 
+  it('withMaxCompletionTokens preserves explicit defaultMaxTokens above the ceiling for known models', async () => {
+    const provider = new AnthropicChatProvider({
+      model: 'claude-opus-4-7',
+      apiKey: 'test-key',
+      stream: false,
+      defaultMaxTokens: 999999,
+    }).withMaxCompletionTokens(1024);
+    const history: Message[] = [
+      { role: 'user', content: [{ type: 'text', text: 'hi' }], toolCalls: [] },
+    ];
+    const body = await captureRequestBody(provider, '', [], history);
+
+    expect(body['max_tokens']).toBe(999999);
+  });
+
   it('withMaxCompletionTokens clamps above the documented ceiling for known models', async () => {
     const provider = new AnthropicChatProvider({
       model: 'claude-opus-4-7',