They are really making a run for it. My understanding is that there was a big push for data center build a few years ago in China but demand didn't rise as expected so there was a lot of unused compute capacity already sitting idle and that's giving these companies super cheap access to readily available compute and they are sure taking advantage of it.
Yep, I made a mistake by using the word "Models" instead of "Labs" or "Model Families", and I can't correct it now. Anyway, it doesn't change the main point. The "Models" rankings are just different Claude variants affecting the results, but the overall picture in terms of model families remains the same.
Are these arena sites gamed? If China and their companies are pushing so hard that they regularly perform distillation and cross other boundaries, why wouldn’t they also try to influence scores here?
Thank you for pointing that out, but there is no error in the sorting. The only mistake is that the title uses the word "Models" instead of "Labs" or "Model Families". Unfortunately, I can no longer correct it, but it doesn't change the substance of the question in any way, since the top models list is made up of models from those companies. There are simply more models because they are available in different sizes (like multiple Claude variants, etc.).
They are really making a run for it. My understanding is that there was a big push for data center build a few years ago in China but demand didn't rise as expected so there was a lot of unused compute capacity already sitting idle and that's giving these companies super cheap access to readily available compute and they are sure taking advantage of it.
This is showing a ranking per lab not model.
If you rank by model only 2 Chinese models are in the top 10.
Yep, I made a mistake by using the word "Models" instead of "Labs" or "Model Families", and I can't correct it now. Anyway, it doesn't change the main point. The "Models" rankings are just different Claude variants affecting the results, but the overall picture in terms of model families remains the same.
Are these arena sites gamed? If China and their companies are pushing so hard that they regularly perform distillation and cross other boundaries, why wouldn’t they also try to influence scores here?
Misinformation. Incorrect sorting selected to prove a point.
Nice try OP.
Thank you for pointing that out, but there is no error in the sorting. The only mistake is that the title uses the word "Models" instead of "Labs" or "Model Families". Unfortunately, I can no longer correct it, but it doesn't change the substance of the question in any way, since the top models list is made up of models from those companies. There are simply more models because they are available in different sizes (like multiple Claude variants, etc.).