Key Info
AutoTrust AI introduced JEV-27B-VL, described as the world's first open-weight near-SOTA multimodal decision model, built with its Blocks of Experts (BoE) recipe, that turns visual state into calibrated action probabilities and selects what to do next.
Highlights
- Claimed as first open-weight near-SOTA multimodal decision model
- Built using AutoTrust AI's Blocks of Experts (BoE) recipe
- Converts visual state into calibrated action probabilities rather than just describing images
- Demo shows it evaluating a Mario-like scenario: position, movement, obstacles, timing
- Decides between actions like running, jumping, or both