diff --git a/docs/cn/ai-providers.md b/docs/cn/ai-providers.md index d963e688..8ab5b07f 100644 --- a/docs/cn/ai-providers.md +++ b/docs/cn/ai-providers.md @@ -149,9 +149,11 @@ AZURE_REASONING_SUMMARY=detailed # 可选：none, brief, detailed AWS_REGION=us-west-2 AWS_ACCESS_KEY_ID=your_access_key_id AWS_SECRET_ACCESS_KEY=your_secret_access_key -AI_MODEL=anthropic.claude-sonnet-4-5-20250514-v1:0 +AI_MODEL=global.anthropic.claude-sonnet-5-5 ``` +较新的 Claude 模型只能通过推理配置文件（inference profile）的 ID 调用，这种 ID 以地区前缀开头，例如 `global.` 或 `us.`。直接用 `anthropic.claude-sonnet-5-5` 这样的 ID 会报错 “on-demand throughput isn't supported”。 + 注意：在 AWS 环境（Lambda、带有 IAM 角色的 EC2）中，凭证会自动从 IAM 角色获取。 ### OpenRouter diff --git a/docs/en/ai-providers.md b/docs/en/ai-providers.md index c23bf9c5..cce04dcd 100644 --- a/docs/en/ai-providers.md +++ b/docs/en/ai-providers.md @@ -164,9 +164,11 @@ AZURE_REASONING_SUMMARY=detailed # Optional: none, brief, detailed AWS_REGION=us-west-2 AWS_ACCESS_KEY_ID=your_access_key_id AWS_SECRET_ACCESS_KEY=your_secret_access_key -AI_MODEL=anthropic.claude-sonnet-4-5-20250514-v1:0 +AI_MODEL=global.anthropic.claude-sonnet-5-5 ``` +Recent Claude models answer only through an inference profile id, which starts with a region prefix such as `global.` or `us.`. The plain id (`anthropic.claude-sonnet-5-5`) fails with "on-demand throughput isn't supported". + Note: On AWS (Lambda, EC2 with IAM role), credentials are automatically obtained from the IAM role. ### OpenRouter diff --git a/docs/ja/ai-providers.md b/docs/ja/ai-providers.md index b09a8bb5..3974a90d 100644 --- a/docs/ja/ai-providers.md +++ b/docs/ja/ai-providers.md @@ -149,9 +149,11 @@ AZURE_REASONING_SUMMARY=detailed # 任意: none, brief, detailed AWS_REGION=us-west-2 AWS_ACCESS_KEY_ID=your_access_key_id AWS_SECRET_ACCESS_KEY=your_secret_access_key -AI_MODEL=anthropic.claude-sonnet-4-5-20250514-v1:0 +AI_MODEL=global.anthropic.claude-sonnet-5-5 ``` +最近の Claude モデルは、推論プロファイル（inference profile）の ID でのみ呼び出せます。この ID は `global.` や `us.` などのリージョン接頭辞で始まります。`anthropic.claude-sonnet-5-5` のような接頭辞なしの ID では「on-demand throughput isn't supported」というエラーになります。 + 注: AWS 上（IAM ロールを持つ Lambda や EC2）では、認証情報は IAM ロールから自動的に取得されます。 ### OpenRouter diff --git a/lib/deprecated-params.ts b/lib/deprecated-params.ts index 612338b2..2d9d2fda 100644 --- a/lib/deprecated-params.ts +++ b/lib/deprecated-params.ts @@ -18,21 +18,25 @@ interface CallParams { providerOptions?: Record | undefined> } -/** Drop a thinking config of type "enabled" stored under key, if any. */ -function withoutEnabledThinking( +// What these models take instead of a budget. Without display "summarized" +// they think but send no thinking text to show. +const ADAPTIVE_THINKING = { type: "adaptive", display: "summarized" } + +/** Turn a thinking config of type "enabled" stored under key into adaptive */ +function adaptiveThinking( options: Record | undefined, key: string, ): Record | undefined { const config = options?.[key] as { type?: string } | undefined if (config?.type !== "enabled") return options - const { [key]: _, ...rest } = options as Record - return rest + return { ...options, [key]: ADAPTIVE_THINKING } } /** * The params without the settings newer Claude models reject, or null when - * the error is about something else or there is nothing to drop. The model - * then runs with its own default sampling and thinking. + * the error is about something else or there is nothing to change. The + * model then runs with its default sampling, and a thinking budget becomes + * adaptive thinking. */ export function withoutDeprecatedParams( error: unknown, @@ -43,8 +47,8 @@ export function withoutDeprecatedParams( const { temperature, topP, topK, ...rest } = params const options = params.providerOptions - const anthropic = withoutEnabledThinking(options?.anthropic, "thinking") - const bedrock = withoutEnabledThinking(options?.bedrock, "reasoningConfig") + const anthropic = adaptiveThinking(options?.anthropic, "thinking") + const bedrock = adaptiveThinking(options?.bedrock, "reasoningConfig") const changed = temperature !== undefined || topP !== undefined || @@ -80,7 +84,7 @@ export function withDeprecatedParamsFallback( const retry = withoutDeprecatedParams(error, params) if (!retry) throw error console.warn( - "[model params] Rejected sampling or thinking settings, retrying with the model defaults", + "[model params] Rejected sampling or thinking settings, retrying with default sampling and adaptive thinking", ) return await inner.doStream(retry) } diff --git a/lib/types/model-config.ts b/lib/types/model-config.ts index f886e736..d8acad4f 100644 --- a/lib/types/model-config.ts +++ b/lib/types/model-config.ts @@ -347,32 +347,35 @@ export const SUGGESTED_MODELS: Partial> = { "o3", "o4-mini", ], + // Newer models only answer through an inference profile id (the region + // prefix). Each id here was called once on 2026-10-04. bedrock: [ - // Anthropic Claude - "anthropic.claude-opus-4-8", - "anthropic.claude-opus-4-7", - "anthropic.claude-sonnet-4-6", - "anthropic.claude-opus-4-6-v1", - "anthropic.claude-opus-4-5-20251101-v1:0", - "anthropic.claude-sonnet-4-5-20250929-v1:0", - "anthropic.claude-haiku-4-5-20251001-v1:0", - "anthropic.claude-opus-4-1-20250805-v1:0", - "anthropic.claude-opus-4-20250514-v1:0", - "anthropic.claude-sonnet-4-20250514-v1:0", - "anthropic.claude-3-5-haiku-20241022-v1:0", + // Anthropic Claude ("global." works from any region) + "global.anthropic.claude-opus-5-5", + "global.anthropic.claude-sonnet-5-5", + "global.anthropic.claude-fable-5-1", + "global.anthropic.claude-opus-5", + "global.anthropic.claude-sonnet-5", + "global.anthropic.claude-opus-4-8", + "global.anthropic.claude-opus-4-7", + "global.anthropic.claude-sonnet-4-6", + "global.anthropic.claude-opus-4-6-v1", + "global.anthropic.claude-opus-4-5-20251101-v1:0", + "global.anthropic.claude-sonnet-4-5-20250929-v1:0", + "global.anthropic.claude-haiku-4-5-20251001-v1:0", + "global.anthropic.claude-sonnet-4-20250514-v1:0", // Amazon Nova - "amazon.nova-2-lite-v1:0", - "amazon.nova-premier-v1:0", + "us.amazon.nova-2-lite-v1:0", "amazon.nova-pro-v1:0", "amazon.nova-lite-v1:0", "amazon.nova-micro-v1:0", // Meta Llama - "meta.llama4-maverick-17b-instruct-v1:0", - "meta.llama4-scout-17b-instruct-v1:0", - "meta.llama3-3-70b-instruct-v1:0", + "us.meta.llama4-maverick-17b-instruct-v1:0", + "us.meta.llama4-scout-17b-instruct-v1:0", + "us.meta.llama3-3-70b-instruct-v1:0", // Mistral "mistral.mistral-large-3-675b-instruct", - "mistral.pixtral-large-2502-v1:0", + "us.mistral.pixtral-large-2502-v1:0", ], openrouter: [ // Anthropic diff --git a/tests/unit/deprecated-params.test.ts b/tests/unit/deprecated-params.test.ts index 13fff590..0331191e 100644 --- a/tests/unit/deprecated-params.test.ts +++ b/tests/unit/deprecated-params.test.ts @@ -31,18 +31,25 @@ describe("withoutDeprecatedParams", () => { }, } - it("drops sampling settings and the thinking budget", () => { + // Measured on Bedrock: Opus 4.7, 4.8 and every Claude 5 model accept + // adaptive thinking, and return its text only with display "summarized" + const ADAPTIVE = { type: "adaptive", display: "summarized" } + + it("drops sampling settings and switches to adaptive thinking", () => { for (const error of [TEMPERATURE, THINKING]) { expect(withoutDeprecatedParams(error, params)).toEqual({ maxOutputTokens: 1000, providerOptions: { - anthropic: { cacheControl: { type: "ephemeral" } }, + anthropic: { + thinking: ADAPTIVE, + cacheControl: { type: "ephemeral" }, + }, }, }) } }) - it("drops a Bedrock thinking budget", () => { + it("switches a Bedrock thinking budget to adaptive thinking", () => { const bedrock = { providerOptions: { bedrock: { @@ -51,7 +58,7 @@ describe("withoutDeprecatedParams", () => { }, } expect(withoutDeprecatedParams(THINKING, bedrock)).toEqual({ - providerOptions: { bedrock: {} }, + providerOptions: { bedrock: { reasoningConfig: ADAPTIVE } }, }) })