autotrust JEV-35B: 35B-A3B open decision model with calibrated probabilities
AutoTrust's open-weights 35B-A3B MoE decision model (Qwen3.5-35B-A3B base, Apache 2.0) that scores up to 256 typed options in one forward pass with calibrated probabilities.
Why it matters
Replaces ranking pipelines with one System-1 decision call at ~204ms per request - Decision Index 0.3 public suite 59.73, ahead of its JEV-27B-VL predecessor in all five benchmark areas.
What you could build with it
A compliance team reviewing thousands of generated documents could route every draft through JEV-35B's typed decisions with calibrated probabilities, auto-approving the confident cases and queuing only uncertain ones for humans. The result: high-throughput document review with a measurable error budget instead of binary guesswork.
Does it hold up?
Early but promising: vendor-run Decision Index 59.73, 51.3% zero-shot on ERQA embodied reasoning with no robot training, 204ms median latency, and a live news-ranking demo; independent third-party evaluations have not appeared yet.
Built with autotrust JEV-35B
- Guru Search - news search and answers ranked by JEV-35B decisionsguru.so · Live, working demo (linked from the model card, launched 2026-10-11) where every news ranking and search result re-rank is a calibrated JEV-35B decision made in about half a second.
- junma/MedJev-Qwen3.5-0.8B - clinical variable extraction as decisionshuggingface.co · Independent third-party JEV-style decision model on Hugging Face: a LoRA decision head on Qwen3.5-0.8B that extracts 11 predefined clinical variables from free-text notes in one pass (~62ms), beating regex and zero-shot baselines.
- autotrust/JEV-35B-NVFP4github · NVFP4-quantized variant of JEV-35B (25.6 GB download, 23 GiB GPU) that runs on a single 32-48 GB GPU with 96.6% answer agreement vs bf16 - linked from the main model card.
Learn more
First spotted on hf: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71JetBrains Mellum 2.1: 12B MoE coding model trained for agentsJetBrains released Mellum 2.1, a 12B MoE model (2.5B active parameters, 128K context) under Apache 2.0, trained mainly…models · JEV 0.69DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.