fix(ai-anthropic): chat({ reasoning }) on Claude 4.6 sends a top-level effort and gets a 400
维护者通常 1 天内回复
已经有一个关联 PR 被合并了。
- #1737 来自 @tombeckenham —— 已合并
评估
- 难度
- 2/5
- 预计耗时
- 1-3 小时
- 新手友好度
- 20/100
- Issue 类型
- 缺陷
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
- 技术栈
- typescript
调研方向
Read packages/ai-anthropic/src/text/reasoning.ts, then inspect the request assertions in packages/ai-anthropic/tests/reasoning-request.test.ts and the Anthropic adapter note in docs/adapters/anthropic.md. Update the Claude 4.6 effort shape, its test, and the documentation as described; done means the tests expect output_config.effort and the docs match. The reporter says they are also opening a PR, so check that work before starting.
由索引模型根据 Issue 内容生成。
描述
TanStack AI version
main at 7532b2b09 (@tanstack/ai 0.68.0, unreleased). The bug came in with #1692 and is not on npm yet. Release PR #1699 is still open.
Framework/Library version
None. This is in @tanstack/ai-anthropic, on the server.
Describe the bug and the steps to reproduce it
chat({ reasoning }) on claude-opus-4-6 or claude-sonnet-4-6 sends the effort as a top-level effort field. Anthropic rejects it with a 400. Both models accept output_config.effort, the same shape the adapter already sends for Claude 4.7 and later.
Every level other than off fails. A call that sets budgetTokens is not affected, because it takes the budget path and sends no effort.
import { chat } from '@tanstack/ai'
import { anthropicText } from '@tanstack/ai-anthropic'
chat({
adapter: anthropicText('claude-sonnet-4-6'),
messages: [{ role: 'user', content: 'Say hi' }],
reasoning: 'low',
})
// 400 invalid_request_error: "effort: Extra inputs are not permitted"
Live results from the Messages API on 2026-10-13, with thinking: { type: 'adaptive', display: 'summarized' }:
| Model | Shape | Result |
|---|---|---|
claude-sonnet-4-6 |
top-level effort: "low" |
400 effort: Extra inputs are not permitted |
claude-sonnet-4-6 |
output_config: { effort: "low" } |
200 |
claude-opus-4-6 |
top-level effort: "low" |
400, same message |
claude-opus-4-6 |
output_config: { effort: "low" } |
200 |
claude-sonnet-4-6 on /v1/messages?beta=true |
top-level effort: "low" |
400, same message |
The adapter sends every call through client.beta.messages, so the last row is the path it uses. That request had no betas header.
Where it is:
packages/ai-anthropic/src/text/reasoning.ts:17definesTOP_LEVEL_EFFORTfor the two 4.6 models. Line 76 returns{ thinking, effort }for them.packages/ai-anthropic/tests/reasoning-request.test.ts:91pins the wrong shape ("Claude 4.6: the effort is a top-level field").docs/adapters/anthropic.md:321says "Claude 4.6 gets adaptive thinking and the top-leveleffort".
Suggested fix: delete TOP_LEVEL_EFFORT and always return output_config: { effort }. Drop the effort? field from AnthropicThinkingFields. Flip the test and fix the docs line.
Your Minimal, Reproducible Example - (Sandbox Highly Recommended)
curl https://api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{"model":"claude-sonnet-4-6","max_tokens":2000,"thinking":{"type":"adaptive","display":"summarized"},"effort":"low","messages":[{"role":"user","content":"Say hi"}]}'
Swap "effort":"low" for "output_config":{"effort":"low"} and the same request returns 200.
Screenshots or Videos (Optional)
No response
Do you intend to try to help solve this bug with your own PR?
Yes, I am also opening a PR that solves the problem along side this issue
Terms & Code of Conduct
- I agree to follow this project's Code of Conduct
- I understand that if my bug cannot be reliable reproduced in a debuggable environment, it will probably not be fixed and this issue may even be closed.
- 主要语言
- TypeScript
- 星标
- 3.2k
- 派生
- 361
- 平均合并
- 1 天 8 小时
- 30 天内合并 PR
- 252
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
TanStack/ai 的其他 Issue
-
难度 4/5 3-5 天 新手友好度 32/100
维护者通常 1 天内回复
-
Responses replay puts assistant text between function_call and function_call_output可能已有人在做 @tombeckenham 今天认领。 未关闭waiting-on: maintainer
维护者通常 1 天内回复
-
move to modelschemas可能已有人在做 @jherr 今天认领。 未关闭has-pr waiting-on: maintainer
维护者通常 1 天内回复
-
Text adapters fail chat()'s adapter constraint under exactOptionalPropertyTypes (midConversationChannels)可能已有人在做 @tombeckenham 今天认领。 未关闭waiting-on: maintainer
维护者通常 1 天内回复
-
chat(): sign the state of a suspended run so a resume without persistence cannot change it可能已有人在做 @AlemTuzlak 于 1 天前认领。 未关闭waiting-on: maintainer
TanStack/ai#1740 · 2 条评论 · 已指派 1 人 ·
维护者通常 1 天内回复
相似的 Issue
-
bug
难度 2/5 1-3 小时 新手友好度 72/100
vercel/streamdown#673 · 1 个 reaction ·
维护者通常 1 天内回复
-
docs(stream-protocol): document messageMetatadata on start/finish and the message-metadata chunk未关闭factory-active task-identify-harness-labels-done task-identify-issue-type-done
难度 2/5 1-3 小时 新手友好度 82/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
agilepathway/label-checker#721 ·
维护者通常 1 天内回复
-
CLI Empathy needs triage
难度 2/5 1-3 小时 新手友好度 74/100
chromaui/chromatic-cli#1524 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
维护者通常 1 天内回复