openclaw - 💡(How to fix) Fix Ollama Cloud models should receive native think: max instead of mapped think: high [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#72402Fetched 2026-04-27 05:30:27
View on GitHub
Comments
1
Participants
2
Timeline
2
Reactions
0
Author
Participants
Timeline (top)
closed ×1commented ×1

The fix in ff570f3 maps /think max to think: "high" for all Ollama models. However, Ollama Cloud endpoints (models with :cloud suffix) accept the native think: "max" value without rejection.

Root Cause

The fix in ff570f3 maps /think max to think: "high" for all Ollama models. However, Ollama Cloud endpoints (models with :cloud suffix) accept the native think: "max" value without rejection.

Code Example

curl -X POST "http://localhost:11434/api/chat" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash:cloud",
    "messages": [{"role": "user", "content": "hi"}],
    "stream": false,
    "options": {"think": "max"}
  }'
RAW_BUFFERClick to expand / collapse

Summary

The fix in ff570f3 maps /think max to think: "high" for all Ollama models. However, Ollama Cloud endpoints (models with :cloud suffix) accept the native think: "max" value without rejection.

Evidence

Live test against ollama/deepseek-v4-flash:cloud (local proxy to Ollama Cloud):

curl -X POST "http://localhost:11434/api/chat" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash:cloud",
    "messages": [{"role": "user", "content": "hi"}],
    "stream": false,
    "options": {"think": "max"}
  }'

Result: Returns thinking content with no errors.

Same endpoint with think: "invalid" also accepted, confirming the Cloud API does not enforce the strict validation that local Ollama does.

Impact

Users on Ollama Cloud models cannot access the full max reasoning effort that the underlying model supports. The mapping to high works but may not be the highest effort level available.

Suggested behavior

  • Local Ollama → keep current mapping: maxhigh
  • Ollama Cloud (:cloud suffix, or detectable via endpoint behavior) → pass think: "max" natively

Alternative: probe think: "max" on first use and fall back to high only if the endpoint rejects it.

Related

  • #71584 (original /think max bug, fixed in ff570f3)
  • #70253 (overlapping streaming config issue)

extent analysis

TL;DR

Update the mapping logic to pass think: "max" natively for Ollama Cloud models while keeping the current mapping for local Ollama models.

Guidance

  • Identify Ollama Cloud models by checking for the :cloud suffix in the model name or by probing the endpoint behavior.
  • Update the code to conditionally map think: "max" to either "high" for local models or pass it natively for Cloud models.
  • Consider implementing a fallback mechanism to handle cases where the endpoint rejects think: "max", such as probing think: "max" on first use and falling back to high if rejected.
  • Verify the updated behavior by testing against both local and Cloud Ollama models using the provided curl command as a reference.

Example

# Example of conditional mapping logic (pseudo-code)
if model_name.endswith(":cloud"):
    think_value = "max"
else:
    think_value = "high"

Notes

The suggested solution relies on accurately detecting Ollama Cloud models, which can be done through the :cloud suffix or endpoint behavior. The fallback mechanism can help handle potential errors or inconsistencies in the Cloud API.

Recommendation

Apply the workaround by updating the mapping logic to conditionally handle think: "max" for Ollama Cloud models, as this allows users to access the full max reasoning effort supported by the underlying model.

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…

Still need to ship something?

×6

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

Back to top recommendations

TRENDING