Open-source Laya takes on Jev: faster and cheaper, but limited label space
Convai released Laya, an open-source non-autoregressive decision model that beats Jev on most benchmarks at 6-8x the speed and runs locally, though Jev still leads on 50+ label spaces while Laya defaults to about 20 options.
Jev just got an open-source rival and it's fast as f**k Convai Innovations dropped Laya - an open-source non-autoregressive decision model that beats Jev on most benchmarks at 6-8x the speed > 421M params, ModernBERT-large backbone, Apache 2.0 > single forward pass in ~33ms on a T4, 100+ languages > runs on less than 1GB, any laptop or phone > $0 self-hosted vs Jev at $0.042/1M tokens > beats Jev on typed-decisions, AG News, DAIR Emotion, and ECE calibration by 3x what you get: - routing across 3 checkpoints (English, multilingual, typed-decisions) in one pip install - mathematically calibrated probabilities trained with RL against proper scoring rules, so it can't bullshit confidence - the real flex: it never generates text, so nothing to parse and nothing to hallucinate important: > ships overconfident out of the refit temperature on your own data first > Jev still wins on 50+ label spaces, Laya tops out around 20 options at defaults > English checkpoint is only 512 context, multilingual goes up to 8k can open source eat Jev's lunch before TypeSafe scales the API?