2026-09-30 · America/Los_Angeles · 社区动态 · #19

Evaluating Open-source Jev alternatives with DecideBench

TL;DR - No open source SystemOne models beat Jev atm based on this benchmark, but they come very close and some are more cost effective. Shoutouts of some models: - Imajev-4B at 95% accuracy - also the cheapest model above 90% - Tev by TogetherAI at 92.8% - Decider-4B and JevK5 at >88% are still worth testing at a lower price and latency level. - Other alternatives like Kev4B, Laya, CLM, Julia-1 still struggle with my production-like dataset. Any other Jev like models I should evaluate in this dataset? Happy to work on it!

热度 45.6 / 100；排名与评分保留该期记录。