🏅 Top traders

Style Control increases the Elo of OpenAI models by ~20 vs Google models.

I don't think style control was the default when the question was created but it is now in their new UI.

Which version is more fair?

Taking a big YES gamble on this being >1500 given the new delay announcement on GPT-5.

Does >1500 elo still count? Like, would something in the 1600s+?

Gpt-4o shot up after getting image generation, beating gpt-4.5. Not sure how chatarena works exactly, but I suspect if gpt-5 comes out without image generation then it might not beat 1500.

@Samaritan 4o got general improvements. I don't think image gen is a factor for the overall leaderboard.

@jim That’s correct, it shot up before the image update, and the image can’t affect it imagefully bc the text based lmarena (the main one) doesnt let the users generate images

@jim Didn't realize that. Thanks for clarifying. Really interesting what they're able to do with the older model. I wonder if they use 4.5 to help refine 4o

I think this is plausible. An intelligent router model alone feels like it should be a +50 elo win. I feel like I'm constantly juggling manually between gpt4.5 and o1-pro these days to get the best results.

1340, from gemini 2.5 pro

u mean 1440 and um technically the highest is bigger than that, it's just dropped recently 🤓

@jim Yeah fair do uk what the highest was?

@Bayesian I'd use 1443

there wont be a gpt 5