A leaderboard
for your leaderboards.

What is it?

Have you ever been sent a benchmark by a customer, or read a competitor’s release blog, and thought to yourself, “Yeah, but that benchmark is total trash,” but didn’t want to seem unprofessional?

Well, now you can simply point them to that benchmark’s position in Leaderboard Arena to implicitly signal just how much of a f-ck you give about GSM8K.

Why?

Working at a frontier model company is exhausting. Customers have no idea how your model works, salespeople have no idea what AI is, and VCs have no idea how stacked your post-training setup is. Instead, they all live and die by their leaderboard of choice.

Customer VP doesn’t like the sound of your TTS generation? No worries! They can easily find a third-rate benchmark that backs them up.

Sales guy getting mogged by your competitors? No problem! Did you know that “on our benchmark of choice, we’re at the Pareto frontier”?

So?

Picking apart these benchmarks is pointless. Just when you realize they were sending emojis to your model and calling it broken, they change the weighting or reset the scores.

Now you don’t need to know anything about benchmarking or evals. Just look at our website and remember:

more = gooder.

Anything else?

Oh yeah, I forgot to mention: we’re retaining 60% of our evals as a “holdout” set to prevent “overfitting.” As if anybody would overfit to our garbage-ass benchmark.