What was actually running
- This FreeToken.link Lab Note records a smoke test performed on August 2, 2026. The tested build was basketikun/chatgpt2api v1.8.0 at commit dc105e51.
- The service was built in Docker, bound to 127.0.0.1:13000, and tested with proxy runtime and clearance automation disabled.
What worked in the bounded test
- An unauthenticated request to /v1/models returned HTTP 401, while an authenticated request returned HTTP 200.
- One imported account was detected as free with active 1, limited 0, abnormal 0, and displayed quota 5.
- One non-streaming auto chat completion returned HTTP 200 in 9.92 seconds with finish_reason stop and total_tokens 42.
What the result does not prove
- This result records one successful smoke test under one local configuration. It does not establish reliability, compliance, stable quotas, safe public exposure, or production suitability.
- The tested route is a reverse-engineered ChatGPT web-session route, not the official OpenAI API.
Questions this lab note should answer
Did the local OpenAI-compatible request work?
Yes. One authenticated, non-streaming completion returned HTTP 200 under the recorded test conditions.
Does this establish a stable free API?
No. A single successful request cannot establish future availability, quotas, reliability, or production fitness.
Was this the official OpenAI API?
No. The tested project wraps a reverse-engineered ChatGPT web-session route behind an OpenAI-compatible interface.
The generator and the evidence boundary are disclosed.
The first draft was generated through the locally hosted chatgpt2api experiment from a fixed list of measured facts. FreeToken then removed no caveats, added no unverified product claims, and manually checked every number against the test record before publication.