Claude fails to return valid structured output

10 posts from 8 people8d activeFading, +0 this week

Posts per day10 posts, Sep 7 to Nov 9

The posts behind it

GitHub issuescrewAIInc/crewAI · alexmarchenko-crewaiOct 27

[BUG] LiteLLM path drops token usage for tool-call turns and structured output (response_model)

🤖 [agent on alexmarchenko-crewai's behalf] ### Description When an `LLM` uses the LiteLLM fallback path, two kinds of call never report token usage. That covers `LLM(model=..., is_litellm=True)` and any model prefix that isn't a native provider, for example `litellm_proxy/...`....

GitHub issuesmlflow/mlflow · NantinaOct 7

[BUG] AI Gateway forwards OpenAI `parallel_tool_calls` into the Anthropic Messages body (Vertex Claude 400s on :rawPredict)

> [!WARNING] > Before submitting a PR, please make sure that: > - A maintainer has triaged this issue and applied the `ready` label > - This issue has no assignee > - No duplicate PR exists > > PRs not meeting these requirements may be automatically closed. ### Issues Policy...

1 reactions1 replies

GitHub issuesBerriAI/litellm · shubhamjain0594Oct 5

[Bug]: Claude structured outputs are dropped after output_config.format is rejected

## Error A JSON-schema request to Claude on Bedrock Invoke (`response_format` or `output_config.format`) fails with: This is returned for Claude models that do not accept native structured output on Invoke, including `bedrock/invoke/global.anthropic.claude-sonnet-5`,...

GitHub issuesvllm-project/vllm · simon-veitner-redhatOct 1

[Bug]: Structured output skips the `<tool_call>` that implicitly ends reasoning, so GLM-4.7/5.x `required`/named tool calls are lost

> **AI disclosure:** this issue was investigated and drafted with an AI assistant (Claude). The reproducers were run on the hardware below, and every number in this issue comes from those runs. A human reviewed the issue before filing. ## Your current environment - vLLM main at...

1 replies

GitHub issueslangchain-ai/langchain · ChampionsZhangSep 23

`ChatAnthropic` sends unsupported tool choice and thinking configurations for Claude Opus 5.5

### Submission checklist - [x] This is a bug, not a usage question. - [x] I added a clear and descriptive title that summarizes this issue. - [x] I used the GitHub search to find a similar question and didn't find it. - [x] I am sure that this is a bug in LangChain rather than...

1 replies

GitHub issuescrewAIInc/crewAI · kimnamuSep 23

[BUG] Bedrock: Claude Opus 5.5 structured output fails with 400 because toolChoice is forced

### Description Thanks for the Bedrock structured-output support from #4280 and the "append `structured_output` without forcing" branch from #4307. With Claude Opus 5.5 on Bedrock, `response_model` / `response_format` fails with a 400 on sync, streaming and async calls. Opus 5...

GitHub issuesBerriAI/litellm · kimnamuSep 23

[Bug]: Bedrock Claude Opus 5.5 returns plain text for response_format json_schema because json_tool_call can't be forced

### Description Thanks for adding Claude Opus 5.5 to the Bedrock cost map on launch day (#42588), and for the Fable 5.1 backstop in #39184 that already keeps these requests clear of Bedrock's forced `tool_choice` 400 One case is left: `response_format` with a `json_schema` on...

GitHub issueslangchain-ai/langchain · ChampionsZhangSep 22

ChatAnthropic.bind_tools() raises ValueError on Anthropic's documented browser_toolset_20260801

2 replies

GitHub issuesrun-llama/llama_index · truecallerabrehamSep 20

[Bug]: StreamingAgentChatResponse silently drops ThinkingBlock and ToolCallBlock when writing to history

### Bug Description In `llama-index-core/llama_index/core/chat_engine/types.py`, `StreamingAgentChatResponse.write_response_to_history` (line 200) and `awrite_response_to_history` (line 259) call `_set_message_content(chat.message, final_text.strip())` before persisting the...

GitHub issuesmodelcontextprotocol/servers · CraigBetxSep 7

server-filesystem: missing inputSchema.type breaks JSON Schema 2020-12 validators (all versions 0.6.2–2025.8.21)

...server to work normally in the main chat window (which tolerates the missing field) but fail to start under the "Cowork and Code sessions" shared-pool context, which validates strictly. Log excerpt: [filesystem] [error] Couldn't start for Cowork and Code sessions. Error: Invalid...

3 replies

Companies and products named

Claude

10 posts

OpenAI

4 posts

DeepSeek

3 posts

LangChain

3 posts

Perplexity

2 posts

Ollama

2 posts

About this problem

Evidence

10 posts from 8 people in 7 places, about 1 a week over 51 days. Mostly on langchain-ai/langchain, crewAIInc/crewAI, BerriAI/litellm. Tools named alongside: OpenAI, DeepSeek, LangChain, Perplexity.

Frustration Frustration 0 of 3· Seen on GitHub issues

How it was grouped

Posts that state a pain and match the "structured-output" rule about Claude. First post Sep 7, 2026, latest Oct 27, 2026. Corroborated: 3 or more posts from 2 or more people or places. Method

The same problem elsewhere

Momentum went from steady to fading.

Nov 3, 2026

Momentum went from fading to steady.

Oct 29, 2026

Momentum went from steady to fading.

Oct 14, 2026

Momentum went from fading to steady.

Oct 7, 2026

Momentum went from rising to fading.

Oct 2, 2026

Momentum went from new to rising.

Sep 25, 2026

First seen with 3 posts.

Sep 25, 2026