Repository navigation
[Bug] OpenCode Go deepseek-v4-pro sends an invalid max_completion_tokens value #37947
Description
Description
When OpenClaw uses the OpenCode Go provider with deepseek-v4-pro, requests consistently fail before a response is generated:
400 max_completion_tokens=384000 cannot be greater than
max_model_len=max_total_tokens=135168. Please request fewer output tokens.
(parameter=max_completion_tokens value=384000)
The request appears to send max_completion_tokens=384000, whereas the upstream endpoint reports a total context limit of 135168 tokens.
Could you please confirm whether deepseek-v4-pro is mapped to the intended model? The reported context limit is incompatible with the configured completion limit and may indicate an incorrect model or context-limit mapping.
Expected behavior: OpenCode should cap max_completion_tokens to a value compatible with the actual model context window, or derive it from the provider's model metadata.
Plugins
OpenClaw integration / OpenCode Go provider.
OpenCode version
Not available from the OpenClaw integration.
Steps to reproduce
- Configure OpenClaw to use the OpenCode Go provider.
- Select
deepseek-v4-pro. - Send a request.
- The provider returns the HTTP 400 error shown above.
Screenshot and/or share link
No share link available. Error output is included above.
Operating System
macOS 15.7.7
Terminal
OpenClaw web interface
Activity
github-actions commentedon Jul 20, 2026
This issue might be a duplicate of existing issues. Please check:
max_completion_tokensderived from a stale context window value)deepseek/deepseek-v4-proreports wrong context size (65536 instead of ~1M), which would cause an over-largemax_completion_tokensto be sentmax_completion_tokensmust be <= model context length (closed)github-actions commentedon Sep 19, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.