According to NanoGPT on X (Nov. 8, 2026), users can now choose their own fallback models on the NanoGPT API. An ordered "models" list can be sent through the Chat Completions, Responses or Messages endpoints. A request can try up to four models if a provider fails before the response starts.
This is the provider's own statement; we have not tested the feature or found an independent report.