No gpt o3

artem

Recent benchmarks and release notes reference an o3 model and it’s top of charts. However, I’m still not seeing it as an option.

Bug or still rolling out?

artem

I’m also seeing both GPT 4o and ChatGPT 4o. Perhaps one of these is o3 and mislabeled?

yiwei-1

artem

No, we haven’t rolled out access to O3 yet; it’s only used for internal benchmarking at the moment.

laiz

Hey! I have pinged the team about this. It could be that the team is first making sure the fair use limits work as intended before exposing the more expensive models.

I’m also seeing both GPT 4o and ChatGPT 4o

These are two separate models. ChatGPT 4o is a bleeding-edge model that receives frequent updates while GPT-4o gets fewer updates and more testing. A truly confusing naming scheme by OpenAI 🙂

batuhan

yiwei-1 Since fair use is coming, could you add the cost to the metadata section (like OpenRouter does)? This way, users wouldn't burn through their credits using the most expensive model for everything; they could test the models and pick the one with the best price/performance ratio.

OpenRouter metadata example

Thibaultmol

batuhan this is planned.
There will also be a mode where The assistant will auto choose the right model.
But before that there should be a feature added that allows people to easily see how $$ each model is in the Assistant UI

@artem fair point, i'll ask to get that corrected.

artem

Makes sense. I was just going off the post:

batuhan

Thibaultmol Will the model router be open source?

Thibaultmol

batuhan I doubt it, probably too specific for kagi's infra that they would open source it