Academind(@academind_real):Hi Jev 👋 1. Yesterday TypeSafe AI launched Jev, a model that returns typed decisions + probabilities, not text. 2. LLMs generate tokens for humans. Jev answers predefined questions (Choice / Score / yes-no) in parallel so software can act on the result with no parsing. 3. Fast, cheap, high quality (from what we see now) ~70–500 ms, $0.042 / million input tokens, output free. Early demos and reviews show frontier-level quality on these decision tasks, with vendor speed/cost claims still needing broader independent tests. Only imagining how this affects and speeds up browser and computer use is absolutely insane.

2026.09.17 13:21

Hi Jev 👋 1. Yesterday TypeSafe AI launched Jev, a model that returns typed decisions + probabilities, not text. 2. LLMs generate tokens for humans. Jev answers predefined questions (Choice / Score / yes-no) in parallel so software can act on the result with no parsing. 3. Fast, cheap, high quality (from what we see now) ~70–500 ms, $0.042 / million input tokens, output free. Early demos and reviews show frontier-level quality on these decision tasks, with vendor speed/cost claims still needing broader independent tests. Only imagining how this affects and speeds up browser and computer use is absolutely insane.