Not winning any acting awards but it feels like that can be fixed!

...I would look at ARC-AGI as they are one benchmark where there has not been significant progress with new model releases—Opus 4 is worse than o3 and Gemini 2.5. I find that to be strong evidence of a slowdown.

If you look at longer-standing benchmarks like mmlu, there's absolutely a huge slowdown. I was listening to the head of product for anthropic yesterday and he swears there's no slowdown and points to all of the new benchmarks that have been invented and beaten. Personally...

Does it play well with the apache ecosystem too? arrow interop would be cool.

Dangerous living for a Saturday morning, Tim!

Yes, the top of the whole feed. for everyone.

I had to scroll down my blue sky feed too far to find this. They should just pin you to the top or something.

Seems useful! I will play with it myself

4.5 would be too touchy-feely to put up a good fight. My money's on o3mini.

14k to have hugging face yell at you about MPS non-stop for another 2 years? No thanks

Straight up yesterday I was searching for the openapi.json for a server I'm using. Github kept servicing references to OpenAI in the same server repo. Super annoying

Oh wow. End of an era.

End of an era I guess. Not bad for their last non-reasoning model.

I'm being very critical. It seems like it's probably a great model. I'll use it a lot.

And that's it? Ok.

Ok, 3x improvement on swe-lancer over o3? Now we're talking