O3 gone from Composer

Why is o3 no longer an option in Composer? Is this a bug or is an update coming?

It was there yesterday but I no longer have access to it. Anyone know why?

2 Likes

have this problem too… You can choose new o3-mini in settings but in fact its not o3 mini - wrote response without thinking… I ask what a model this - its say that 3.5 turbo… Resetting the cursor doesn’t help. Developers explain about o3 mini, what is wrong?

Hey, is this model missing for you? Or does this model not work with Composer? What version of Cursor are you using? I just checked, and everything works for me.


2 days ago after updating from version 0.45.11 to 0.45.14 and now 0.45.15, the model that was already selected automatically disappeared… In the models I found it, but it became located at the end… I turned it on and now it does not answer as o3 but 3.5 turbo model - that is instantly any answer gives that in chat mode that in agent mode - that is not thinking at all and instantly gives answers and not qualitatively absolutely - it is clear that it is not o3 mini.

Deepseek-r1 also now fails to show reasoning and instantly writes a crooked answer - clearly not Deepseek-r1… It seems that instead of thinking models in chat mode and which are available in agent mode a completely different model works

I see. No wonder the models are so strange these days.They changed the model because it was really strange, O3’s thinking speed was super fast, which was not like the normal O3,My integrity towards cursor has declined a bit. :slightly_frowning_face:

Prior to these theories, did you try going to to the real ChatGPT and asking the same question?

Anyway, at least for me in Cursor it answers 4 (similar to the original ChatGPT’s answer) and not 3.5…and it takes it time to answer.


It also doesn’t know what is even o3-mini:

But at least DeekSeek-r1 does know itself:

I messaged support and they fixed it immediately.

1 Like

Fixed how?

As said multiple times, models don’t actually know who they are, this is probably done on purpose to offer flexibility about choosing who they are, Deepseek knows about itself because an instruction like “You’re deepseek R1” is being given before prompts.

I had to email support and they fixed it immediately.
They made a change on the backend and then o3 mini was back in my list. Not sure what they did, but Claude 3.7 is much better now.

  1. I created this topic, two, the model list. I was unable to select 03 from the model list, which was fixed by support almost immediately.

Oh, okay, but I assume the model still answers incorrectly about itself as mentioned above.