Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

I get results with several [object Object] in the reply with Mistral, in session files and memory.

I'm sorry to bother you with this issue.
This is exactly issue is the same as #70806
It was suppose to be corrected one or two versions but still persists.

The problem, I guess, is the json from of the reply from Mistral.
I paste the URL of the documentation.

Steps to reproduce

1. Update openclaw 2026.4.27

2. Send a /new

3. Set up a mistral model WITH thinking (>= medium)

4. Send a "Hello"

5. You get replies WITH [object Object]

6. Set up a mistral model WITHOUT thinking or another provider and model (ex: ollama + gemma4)

7. Send a "Hello"

8. You get replies WITHOUT [object Object]

Expected behavior

Clean replay without [object Object]

Actual behavior

reply with leading [object Object]

OpenClaw version

2026.4.27

Operating system

MacOS tahoe

Install method

npm

Model

mistral-small and mistral-large

Provider / routing chain

openclaw -> mistral

Additional provider/model setup details

No response

Logs, screenshots, and evidence

<img width="1276" height="856" alt="Image" src="https://github.com/user-attachments/assets/a5226214-8a71-478f-8dfe-1af553f2e605" />

Impact and severity

Affected all channels
Severity : annoying and data
Frequency : always with thinking
Consequence : lower memory quality, extra tokens

Additional information

The mistral documentation.
https://docs.mistral.ai/studio-api/conversations/reasoning/native