Issue Station
Sign In
Tools hereToolbox

openclaw - 💡(How to fix) Fix [Bug] MiniMax-M2.7 streaming not working - content contains thinking tags [1 comments, 1 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#59823•Fetched 2026-04-08 02:40:09
View on GitHub
Comments
1
Participants
1
Timeline
3
Reactions
0
Author
zcya198zcya198
Participants
zcya198zcya198
Timeline (top)
closed ×1commented ×1locked ×1

Root Cause

Compared SSE responses between GLM-5 (working) and MiniMax-M2.7 (broken):

Fix Action

Workaround

Use dashscope/MiniMax-M2.5 (alias: MiniMax-Dash) instead, which does not have this issue.

RAW_BUFFERClick to expand / collapse

Bug Description

Problem: Streaming output does not work with minimax-cn/MiniMax-M2.7 model in Feishu channel. The response appears all at once instead of streaming token by token.

Expected: Streaming should work the same as GLM-5 (dashscope) which streams correctly.

Actual: minimax-cn/MiniMax-M2.7 shows no streaming, waits for complete response before displaying.

Environment

  • OpenClaw version: 2026.4.1 (da64a97)
  • Channel: Feishu
  • Model: minimax-cn/MiniMax-M2.7

Root Cause Analysis

Compared SSE responses between GLM-5 (working) and MiniMax-M2.7 (broken):

GLM-5 (dashscope) - Working

  • content is null
  • reasoning_content contains thinking
  • Clean separation

MiniMax-M2.7 (minimax-cn) - Broken

  • content contains BOTH thinking tags (think) AND actual response text mixed together
  • Content example: "<think>\nThe user says "hi"."
  • Extra fields: name, audio_content, input_sensitive, output_sensitive, etc.

Hypothesis

OpenClaw SSE streaming parser:

  1. Reads delta.content and displays it when non-null
  2. Does not understand the (think) thinking tags embedded in the content
  3. Treats "(think)\nThe user..." as plain text and fails to process correctly for streaming display

The thinking tags in the content field confuse the streaming display logic.

Workaround

Use dashscope/MiniMax-M2.5 (alias: MiniMax-Dash) instead, which does not have this issue.

Config Changes Tried

  1. Changed minimax-cn baseUrl from /anthropic to /v1
  2. Changed minimax-cn api from anthropic-messages to openai-completions
  3. Added no_proxy for api.minimaxi.com

None of these fixed the streaming issue. The root cause is in how OpenClaw parses the SSE content field with thinking tags.

Expected Behavior

MiniMax-M2.7 streaming should work like GLM-5, with tokens appearing progressively in the chat.

User: 骆伟雄 (ou_97861c3cd1c00cff9d8002f2e68bdb94) Reported via: OpenClaw ops agent

extent analysis

TL;DR

The most likely fix is to modify the OpenClaw SSE streaming parser to correctly handle the thinking tags embedded in the content field of the MiniMax-M2.7 model's responses.

Guidance

  • Verify that the issue is indeed caused by the thinking tags in the content field by checking the SSE responses from the MiniMax-M2.7 model.
  • Consider using the dashscope/MiniMax-M2.5 model as a temporary workaround, as it does not have this issue.
  • To fix the issue, the OpenClaw SSE streaming parser needs to be updated to parse the content field correctly, potentially by ignoring or stripping the thinking tags.
  • Test the updated parser with the MiniMax-M2.7 model to ensure that streaming works as expected.

Example

No code example is provided as the issue does not include specific code snippets or APIs.

Notes

The provided workaround using the dashscope/MiniMax-M2.5 model may not be a long-term solution, and updating the OpenClaw SSE streaming parser is likely necessary for a permanent fix.

Recommendation

Apply the workaround by using the dashscope/MiniMax-M2.5 model until the OpenClaw SSE streaming parser can be updated to correctly handle the thinking tags in the MiniMax-M2.7 model's responses.

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#device allocation#model download#tokenizer error#prompt formatting

Still need to ship something?

×6

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

Back to top recommendations

TRENDING