Expose reasoning effort capability in the OpenAI-compatible API

#2
by lhfe - opened

I am serving Qwen3.8-27B-Escha-W2 with the official Escha runtime.

The model chat template supports configurable reasoning effort:

  • xhigh (default)
  • medium
  • low

However, the OpenAI-compatible GET /v1/models endpoint currently returns only:

{
  "id": "escha-qwen38-27b-w2",
  "object": "model",
  "owned_by": "sglang",
  "max_model_len": 262144
}

Sign up or log in to comment