Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

When ollama/minimax-m3:cloud is only added via agents.defaults.models, OpenClaw’s running gateway can select and run the model but does not expose its Ollama-reported thinking capability unless explicit models.providers.ollama.models[] metadata is added.

Steps to reproduce

Shortest observed repro:

# 1. Allow/select the live-discovered model, but do not add a models.providers.ollama.models[] row for it
openclaw config set agents.defaults.models '{"ollama/minimax-m3:cloud":{}}' --strict-json --merge
openclaw gateway restart

# 2. Ollama itself reports thinking support
curl -s http://127.0.0.1:11434/api/show \
  -H 'Content-Type: application/json' \
  -d '{"name":"minimax-m3:cloud"}'

# 3. Local OpenClaw path accepts thinking
openclaw infer model run --local \
  --model ollama/minimax-m3:cloud \
  --thinking low \
  --prompt "Reply with exactly: pong" \
  --json

# 4. Gateway path rejects the same model/thinking level
openclaw infer model run --gateway \
  --model ollama/minimax-m3:cloud \
  --thinking low \
  --prompt "Reply with exactly: pong" \
  --json

Observed failure: step 4 returns Thinking level "low" is not supported for ollama/minimax-m3:cloud. Use one of: off.

Expected behavior

Expected result, based on observed local behavior: because Ollama /api/show reports thinking for minimax-m3:cloud and openclaw infer model run --local --model ollama/minimax-m3:cloud --thinking low succeeds, the gateway path should also accept --thinking low for the same ollama/minimax-m3:cloud model without requiring a manual models.providers.ollama.models[] metadata entry.

Actual behavior

Observed result: curl /api/show for minimax-m3:cloud reports capabilities including thinking, and openclaw infer model run --local --model ollama/minimax-m3:cloud --thinking low returns ok: true, but the equivalent gateway command fails with GatewayClientRequestError: Error: Thinking level "low" is not supported for ollama/minimax-m3:cloud. Use one of: off.; after adding an explicit models.providers.ollama.models[] row with reasoning: true and restarting the gateway, the same gateway command returns ok: true.

OpenClaw version

2026.6.1

Operating system

Debian 13

Install method

npm global

Model

minimax-m3:cloud (Ollama Cloud via local 127.0.0.1:11434 in docker)

Provider / routing chain

openclaw -> local ollama docker instance -> ollama cloud -> minimax-m3

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response