openclaw - 💡(How to fix) Fix [Bug]: DeepSeek V4-pro 400 error via opencode-go: reasoning_content passthrough regression in v2026.4.26 [2 comments, 3 participants]

Error Log

'13:46:46 [agent/embedded] embedded run failover decision: runId=52c61bf4-99d1-4874-bbb8-efa9188228eb stage=assistant decision=surface_error reason=format from=opencode-go/deepseek-v4-flash profile=sha256:0c6f568f3b65 rawError=400 Error from provider (DeepSeek): The reasoning_content in the thinking mode must be passed back to the API.'

TL;DR

The issue can likely be fixed by modifying OpenClaw to include the previous reasoning_content in the history payload for multi-turn conversations with DeepSeek V4 models.

Guidance

• Verify that the reasoning_content is being returned by the provider in the first turn and check how it's being handled in the OpenClaw code.
• Investigate the previous reasoning_content preservation logic to see if it's incomplete or if there's a regression that's causing the issue.
• Check the OpenClaw documentation to see if there are any configuration options or workarounds for preserving the reasoning_content in multi-turn conversations.
• Test the issue with different models (e.g., deepseek-v4-pro vs deepseek-v4-flash) to see if the problem is specific to one model.

Example

No code snippet is provided as the issue doesn't contain enough information about the specific code changes required.

Notes

The issue seems to be related to a regression or incomplete fix in the OpenClaw code, and the solution will likely involve modifying the code to properly handle the reasoning_content in multi-turn conversations.

Recommendation

Apply a workaround by modifying the OpenClaw code to include the previous reasoning_content in the history payload, as this is the most direct way to address the issue.