Issue Station
Sign In
Tools hereToolbox

openclaw - 💡(How to fix) Fix DeepSeek: thinking mode still broken in v4.24 (content[].thinking must be passed back) [1 comments, 2 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
openclaw/openclaw#71998•Fetched 2026-04-27 05:36:18
View on GitHub
Comments
1
Participants
2
Timeline
4
Reactions
0
Author
CSZ-wCSZ-w
Participants
CSZ-wCSZ-wsteipetesteipete
Timeline (top)
cross-referenced ×2closed ×1commented ×1

Error Message

n14:06:32 [agent/embedded] embedded run agent end: runId=56761779-775d-4e6d-a434-0ece2d2e3366 isError=true model=deepseek-v4-pro provider=deepseek error=LLM request failed: provider rejected the request schema or tool payload. rawError=400 {"error":{"message":"The content[].thinking in the thinking mode must be passed back to the API.","type":"invalid_request_error","param":null,"code":"invalid_request_error"}} n

RAW_BUFFERClick to expand / collapse

Bug description

OpenClaw 2026.4.24 changelog states:

DeepSeek thinking/replay behavior is fixed for follow-up tool-call turns.

However, the content[].thinking passing error persists. When DeepSeek V4 Pro (or V4 Flash) is used with think mode enabled, the provider rejects the request with a 400 error, causing a silent fallback chain.

Environment

  • OpenClaw: 2026.4.24 (cbcfdf6), upgraded from 2026.4.15 (041266a)
  • Model: deepseek/deepseek-v4-pro (via nthropic-messages API)
  • Base URL: https://api.deepseek.com/anthropic
  • Node.js: v24.14.0
  • OS: Windows 11 24H2 (build 26200, x64)
  • Think: enabled (triggers bug) / disabled (works fine)
  • Reasoning: enabled (does not affect the bug)

Error

n14:06:32 [agent/embedded] embedded run agent end: runId=56761779-775d-4e6d-a434-0ece2d2e3366 isError=true model=deepseek-v4-pro provider=deepseek error=LLM request failed: provider rejected the request schema or tool payload. rawError=400 {"error":{"message":"The content[].thinking in the thinking mode must be passed back to the API.","type":"invalid_request_error","param":null,"code":"invalid_request_error"}} n

Fallback chain

n14:06:33 deepseek/deepseek-v4-pro ? fail (400, invalid_request_error) 14:06:33 moonshot-coding-plan/kimi-for-coding ? fail (402, membership inactive) 14:06:44 minimax-token-plan/MiniMax-M2.7 ? success (silent fallback) n

Reproduction

  1. Set model to deepseek/deepseek-v4-pro`n2. Enable think mode
  2. Send any message
  3. Observe: 400 error in logs, silent fallback to minimax
  4. Disable think: works normally

Notes

  • Same error existed in v4.15, confirmed still present in v4.24
  • The 4.24 changelog fix likely addressed a different aspect (possibly replay/idempotency) but not the initial content[].thinking schema issue
  • OpenClaw sends Anthropic-format thinking blocks to the DeepSeek API, but DeepSeek expects its own thinking format or requires that thinking blocks from prior turns be returned verbatim

Related

  • #71976 (also reported by same user)

extent analysis

TL;DR

The issue is likely due to a mismatch between the thinking format sent by OpenClaw and the expected format by the DeepSeek API, causing a 400 error when think mode is enabled.

Guidance

  • Verify that the content[].thinking field is being passed correctly in the API request to DeepSeek, and that it matches the expected format.
  • Check the OpenClaw documentation to see if there are any configuration options or workarounds for handling the thinking format difference between OpenClaw and DeepSeek.
  • Test the API request with the think mode disabled to confirm that the issue is specific to the thinking mode.
  • Review the related issue #71976 to see if there are any additional insights or potential solutions.

Example

No code snippet is provided as the issue is related to the interaction between OpenClaw and the DeepSeek API, and the exact code changes would depend on the specific implementation.

Notes

The issue may be specific to the DeepSeek V4 Pro model, and the fix may require changes to the OpenClaw configuration or the API request format. The fact that the issue persists across different OpenClaw versions (2026.4.15 and 2026.4.24) suggests that the problem is not related to a specific version.

Recommendation

Apply a workaround to handle the thinking format difference between OpenClaw and DeepSeek, such as modifying the API request to include the expected thinking format or using a different model that does not require this specific format.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…
#api#LLM response#prompt template#agent execution#callback error

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING