AutoTrust AI debuts JEV-27B-VL, an open-weight multimodal decision model that acts on what it sees
alexcovo_eth · x · 2026-10-01
AutoTrust AI introduced JEV-27B-VL, which it claims is the world's first open-weight, near-SOTA multimodal decision model, built with its Blocks of Experts (BoE) recipe.
Instead of merely describing images, the model converts visual state into calibrated action probabilities and then picks the next action — a see → decide → act loop. In a demo, it tracks Mario's position, movement, obstacles, and timing in Super Mario Bros., choosing whether to run, jump, or both.
More from Multimodal
- Single-word prompt experiment: generating layered imagery with "Palimpsest" — tisch_eins · 2026-10-01
- Hands-On With LTX-2.5 SDR-to-HDR IC-LoRA: 1-6% of Pixels Exceed SDR Range — Liquidrider · 2026-10-01
- Suno Studio 2.0 adds chat-based stem splitting, instruments, effects and automation — suno · 2026-10-01
- AI short film "We Stole a UFO" puts a grey alien on galactic trial — Puzzleheaded-Mall528 · 2026-10-01
- Mixie3D launches: AI agents model, texture and light 3D scenes in parallel — gowthami_s · 2026-10-01
- A Qwen outpaint LoRA beats native outpainting on quality and consistency — linoy_tsaban · 2026-10-01