Request contract
POST /v1/chat/completions accepts JSON messages, model, max_tokens or max_completion_tokens, temperature, top_p, stop, and stream. Unknown or unbounded model/output values are rejected before forwarding.
json
{"model":"qwen3.8-27b","messages":[{"role":"user","content":"Summarize this text."}],"temperature":0.2,"max_tokens":256,"stream":false}Response and usage
Non-streaming responses use the OpenAI completion envelope. Usage contains prompt, cached prompt, completion, and total token counts when the backend provides authoritative usage. The settled debit is rounded once per request.
Reference model
qwen3.8-27b