AutoTrust 发布开源决策模型 JEV-27B，性能略超闭源 Jev 1.13
原标题：autotrust/JEV-27B: fast, calibrated decisions and full reasoning from one open model
AI 摘要
AutoTrust AI 发布开源权重模型 autotrust/JEV-27B（Apache-2.0），在六个公开决策基准上平均得分 84.07，比闭源托管模型 TypeSafe Jev 1.13 高 0.22 个百分点。该模型采用「专家块」方案，将 Qwen3.8-27B 主干冻结，仅训练 1.089 亿参数的 System 1 决策块，同时保留原模型的生成与推理能力（HumanEval 78.0% 不变）。模型可在单张 H100 上以 bf16 全精度运行，每秒约 100 次决策，训练成本约 9.2 个 B200 小时。
正文节选
Today we are introducing autotrust/JEV-27B, an open-weights (Apache-2.0) model that answers these questions with calibrated probabilities. In our runs it averages 0.22 percentage points higher than TypeSafe Jev 1.13 on six public decision benchmarks. The same weights also serve an untouched Qwen3.8-27B for ordinary generation and step-by-step reasoning (System 2), from one vLLM engine. - Six public decision benchmarks, one protocol: JEV-27B averages 84.07, TypeSafe Jev 1.13 83.85 in our runs of