openclaw - 💡(How to fix) Fix NVIDIA API doNVIDIA custom provider hangs with thinking=medium — no way to disable per provideresn't support it. Need a way to disable thinking per custom provider. [2 comments, 2 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

Custom NVIDIA provider freezes when OpenClaw auto-sets thinking=medium because NVIDIA API does not support this parameter.

Steps to reproduce

1. Add custom provider: custom-integrate-api-nvidia-com/meta/llama-3.1-70b-instruct
2. Set it as agents.defaults.model.primary
3. Send any message via Telegram
4. Observe bot returns cached "Standing by" instead of live response

Expected behavior

Bot responds normally. thinking parameter should not be sent to providers that don't support it.

Actual behavior

Gateway sets thinking=medium for NVIDIA model. NVIDIA API does not support this parameter and hangs. Bot returns 100% cached response.

OpenClaw version

2026.5.6

Operating system

Windows 11 + WSL2 Ubuntu

Install method

npm global

Model

custom-integrate-api-nvidia-com/meta/llama-3.1-70b-instruct

Provider / routing chain

openclaw -> NVIDIA integrate AP

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

Severity: High — blocks all responses Frequency: Always Consequence: Bot unusable with NVIDIA custom provider as primary model Workaround: Use google/gemini-2.5-flash as primary instead

Additional information

No response