Maincode ships Matilda Jev, an open ~26B decision model that claims a Decision Index of 59.59
Maincode published Matilda Jev on Hugging Face on 30 September 2026: an open ~26.1B System One model with a 255-option readout, a stored calibration temperature, choice / noul / score questions and optional images. It ships its own server for POST /v1/systemone, so clients written for Jev can point at it. The license is Apache-2.0.
The headline number (UnverifiedClaim). The launch card reported Decision Index 0.2.1 59.26 over 150,317 requests. On 2 October (00:01 UTC) Maincode pushed v1.3 weights and the card now says 59.59 (raw 69.18, breadth 58.39). For reference, Jev 1.13 sits at 57.91 on the official board. The card says the model is “currently awaiting official submission”, and the Decision Index board had no Matilda row when we checked on 2 Oct. By area it’s strongest on tools & automation and weakest on knowledge & reasoning.
How we list it. Open trained decision weights plus a typed probability API → catalog at 5.4 / 6.5 / 5. Capability is held back by the missing latency and calibration numbers and by the self-run Index; adoption is day-two.