Reasoning Effort

[dusty ocean]

Using Kimi K2.5 on OpenClaw and this apppears before every response. The response are fine, but the error keeps appearing. I have reasoning and thinking disabled.

HTTP 400: We got a bad response from the source. Status 400. Error message: {"error":{"message":"thinking is enabled but reasoning_content is missing in assistant tool call message at index 9","type":"invalid_request_error"}}

[ruby lotus]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[flat cairn]

That 400 is coming from Kimi’s OpenAI-compatible API: it thinks “thinking” is enabled, so it requires a reasoning_content field on every assistant message in the request history — including assistant tool_call messages. OpenClaw’s tool_call messages don’t include that field, so the provider rejects the request… and OpenClaw then (apparently) retries in a way that still produces a normal answer, which is why you see the error but also get a response.

A few practical fixes/workarounds:

1. Double-check you’re not on a “thinking” Kimi model

• If you’re using moonshot/kimi-k2-thinking or moonshot/kimi-k2-thinking-turbo, switch to the non-thinking model:
• openclaw models set moonshot/kimi-k2.5
  Docs: https://docs.openclaw.ai/providers/moonshot

2. Start a fresh session after switching
   Even if you switch models / toggle things, the existing conversation history can keep triggering the provider’s validation. New session = clean history.

3. Verify your session thinking level (separate from “reasoning visibility”)
   In OpenClaw, /reasoning off just hides reasoning text; it doesn’t necessarily stop a reasoning-capable model mode.

• Send /think to see what the session is set to, and /think:off to force it off.
  Docs: https://docs.openclaw.ai/tools/thinking