#400 Item '' of type 'reasoning' was provided without its required following item.
1 messages · Page 1 of 1 (latest)
That error (400 Item 'rs_…' of type 'reasoning' was provided without its required following item) is coming from OpenAI’s Responses API. It usually means your session history contains an “orphaned” reasoning block (often happens if a turn got interrupted/aborted mid-stream), and then the next user message replays that broken history → OpenAI rejects the request (so it looks like “it fails on the 2nd message”).
Quick workarounds
- In that same chat, send
/reset(or/new) as a standalone message, then try again. - If you ever used
/stop(or the gateway restarted) while GPT‑5 was “thinking”, avoid doing that for now. - If you just need it to run reliably: send
/think offfor that chat (disables reasoning blocks).
If you want to help confirm what case you’re hitting
Paste:
openclaw --version- the exact model id you picked (e.g.
openai/gpt-5.2-chat-latest) - the error context from
openclaw logs --followright when it fails
This is a known issue around 2026.2.15 + GPT‑5.x (OpenAI reasoning replay). Tracking issues:
2026.2.15, gpt-5
21:12:57 debug agent/embedded {"subsystem":"agent/embedded"} embedded run start: runId=3744155c-7644-4636-aa19-51fad5c2b94e sessionId=0aca9012-7864-4575-a962-d2dbd6d24339 provider=openai model=gpt-5 thinking=low messageChannel=webchat
21:12:57 debug diagnostic {"subsystem":"diagnostic"} run registered: sessionId=0aca9012-7864-4575-a962-d2dbd6d24339 totalActive=1
21:12:57 debug agent/embedded {"subsystem":"agent/embedded"} embedded run prompt start: runId=3744155c-7644-4636-aa19-51fad5c2b94e sessionId=0aca9012-7864-4575-a962-d2dbd6d24339
21:12:57 debug agent/embedded {"subsystem":"agent/embedded"} embedded run agent start: runId=3744155c-7644-4636-aa19-51fad5c2b94e
21:12:58 debug agent/embedded {"subsystem":"agent/embedded"} embedded run agent end: runId=3744155c-7644-4636-aa19-51fad5c2b94e isError=true
21:12:58 debug agent/embedded {"subsystem":"agent/embedded"} embedded run prompt end: runId=3744155c-7644-4636-aa19-51fad5c2b94e sessionId=0aca9012-7864-4575-a962-d2dbd6d24339 durationMs=788
Yeah — 2026.2.15 + “gpt‑5” (OpenAI Responses) can hit this when the session transcript ends up with an orphaned reasoning item in history. Then the next message replays that history and OpenAI rejects it → so it looks like “it breaks on the 2nd message”.
Fastest workaround
In the same chat, send /reset (or /new) as a standalone message, then try again.
Two questions (to confirm which variant you’re hitting)
- What’s the exact model id shown by
/model status? (e.g. it’ll show the fullopenai/...string, not just “gpt-5”) - Did you use
/stopor otherwise interrupt the first reply before it finished? (that’s the most common way to create an orphaned reasoning block)
If it still happens immediately after /reset
Please paste the relevant lines from:
openclaw logs --follow(the chunk that includes the 400), and- the output of
openclaw status
There are active bug reports for this family of errors; if your case reproduces without /stop, it’d be great evidence to add:
https://github.com/openclaw/openclaw/issues/18282