Why does chat completions report zero tokens

No inference runs. It returns a well-formed response so a client finishes its handshake, with usage zeroed.

Run it

curl -s -X POST https://api.model.tools/v1/chat/completions -H 'Content-Type: application/json' -d '{"model":"any","messages":[]}'

Response

{"usage":{"prompt_tokens":0,"completion_tokens":0}}