My ft:gpt-4.1-nano:xxx is working. So not 100% “down”.
Something that OpenAI has been screwing with is making even fine-tuning models “scoped” in some manner - I just tried inference with another project and that worked, yet that project doesn’t get models in the fine-tuning jobs UI.
Hey @mani.doraisamy , Would you be able to share a request_id with us to identify why your ft models are throwing this error. Since this is not a widespread issue at the moment. Thank you!
But the fine-tuned model still throws “Model not found” in both the Playground and the API.
So the model exists (fine-tune job succeeded and ID is visible), but the inference router can’t load the fine-tuned version. This happens for all gpt-4.1-nano fine-tunes.
Let me know if you need request_ids from the failing calls — I can provide those too.
@Prashant_Pardesi Finetuning on gpt-4.1-mini works fine. We’ve started migrating our nano fine-tunes to mini just to keep production running, but we have many nano models in use, so migrating everything will take time. Before we go down that path, could you please let us know how long the fix for gpt-4.1-nano fine-tuned models is expected to take? It would help us decide whether to proceed with migrating all nano models, or wait for a fix on your side.
Hi @mani.doraisamy , I have had the same problem and the issue is with the Responses API.
If you change to the completions API it will work. I`m not sure why + it is a product bug.
On the playground it is a simple toggle and you can make it work right away.
On your code you will have to make a few adjustments depending on your config. Pay special attention to the json schema in case you are using it as it is a bit different.
We’re experiencing the same thing. Over the last 3 days with 4.1-nano FTs on the Responses API:
11/17/25 - we had 108k FT requests and 107 errors
11/18/25 - we had 96k FT requests and 32k were errors
11/19/25 - we have 73k FT requests and 57k are errors
Tested and resolved by using Completions API.
Hey OpenAI if you’re listening. This is the 2nd time we’ve been burned by sudden severe degraded service on fine tune models. The 1st time it was because the Completions API with gpt-3.5 fine tunes suddenly took 5 seconds to respond at minimum. I would imagine customers with FTs with a sizable frequency of requests are probably core workflows for their customers. It’d be nice if you could at least monitor FT response quality and add to status.openai.com. Thank you for listening.
