updated Decision Index 0.1 → 0.2 🎯 better formula, +29 jev-like models, +21 benchmarks AutoJev-27B by Perplexity CTO Denis Yarats took the open lead, trailing jev by 0.8 points 🏆 come find the best model at every size, speed and use-case

top 1

@perplexity_ai @denisyarats How can we add our models here?

@perplexity_ai @denisyarats do you have a link to the repo?

@perplexity_ai @denisyarats No, but we can give access to the api

@perplexity_ai @denisyarats How do we get listed?

@perplexity_ai @denisyarats send me a github or hugging face link of the model and i'll include it

@perplexity_ai @denisyarats

@perplexity_ai @denisyarats We also have lower context & rate limited public endpoint - if the eval fits

@perplexity_ai @denisyarats DM!

@perplexity_ai @denisyarats @benchmarkheaven add AutoJev-27B

@perplexity_ai @denisyarats Can you add my model

@perplexity_ai @denisyarats Lfg

@perplexity_ai @denisyarats What's truly impressive is how well Jev is holding up. It's obviously really well done because they are able to serve it so cheap and fast, and not even Qwen3.8-27B can beat it in aggregate.

@perplexity_ai @denisyarats yea i think it's a larger pre-trained base model than any of the contenders there's a big model smell in the knowledge benchmarks

@perplexity_ai @denisyarats Yeah, I agree. I think a lot of the open contenders are completely ignoring the fact that when it comes to decisions, the speed is irrelevant if the decision doesn't have proper base and, more importantly, nuance. I wonder how those "it's just an encoder" folks are doing.

@perplexity_ai @denisyarats You should test sage by @levantolabs We're consistently performing best on accuracy

@perplexity_ai @denisyarats Is there anAPI to fetch the leaderboard?

@perplexity_ai @denisyarats Hi i love the animation of this video, is it opensource on how you made this?

@perplexity_ai @denisyarats 0.1 → 0.2 moved models that didn't change at all. if you route on this, pin the index version in the decision record, or 'why did we move off X' becomes unanswerable the next time the formula lands.