Jev launches: a System One model for decisions
TypeSafe founder Diogo Almeida introduces Jev, its RLCD training method, and its focus on software decisions. The post includes a 2:56 launch video.

CLM-8B is out, using System One style with states→actions instead of next-token chat. Claims up to 9× faster than Jev with comparable computer-use/gaming/tools; light FT yields DeepSWE 81.6%, Terminal-Bench 2.1 87.6%. CTO gate before swapping Jev.
CLM-8B just dropped — Contrastive Language Model (System One style): states→actions, not next-token chat. Claims up to 9× faster than Jev with comparable computer-use / gaming / tools; light FT → DeepSWE 81.6%, Terminal-Bench 2.1 87.6%. CTO gate before you swap Jev