Provider API: reasoning_effort "none" is rejected with 400; "off" fails on models that can't disable thinking #985

Summary

On /provider/v1/chat/completions, reasoning_effort: "none", the value OpenAI's API uses to turn reasoning off, is rejected with a 400 on every model I tried. The Command Code-specific value "off" works only on models that can disable thinking; on the others it is also a 400. A standard OpenAI client therefore has no reasoning-off value that works across the catalog.

Possibly related to #697, which was closed after a fix, but "none" is still rejected.

Expected Behavior

• reasoning_effort: "none" is accepted as an alias for "off".
• On models where thinking can't be turned off, "none"/"off" don't hard-fail. They fall back to the lowest supported effort or are ignored, so a client asking for "as little reasoning as possible" gets a response instead of a 400.

Actual Behavior

"none":

{"type":"invalid_request_error","message":"Invalid option: expected one of \"off\"|\"low\"|\"medium\"|\"high\"|\"xhigh\"|\"max\"","param":"reasoning_effort"}

"off" on a model that can't disable thinking:

{"type":"invalid_request_error","code":"unsupported_value","message":"Model \"z-ai/glm-5.3-flash\" does not support reasoning_effort \"off\". Only models that can turn thinking off accept it.","param":"reasoning_effort"}

Steps to reproduce the issue

curl https://api.commandcode.ai/provider/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"z-ai/glm-5.3-flash","reasoning_effort":"none","max_tokens":16,
       "messages":[{"role":"user","content":"Reply OK"}]}'

Swap the model and "none"/"off" to reproduce the table.

Command Code Version

N/A (Provider API, direct HTTP)

Operating System

Linux