Reflection 发布 Beam：中国以外最强开放权重模型
原标题：Reflection's Beam becomes the most capable open-weight model built outside China
AI 摘要
AI 初创公司 Reflection 发布其首个开放权重模型 Beam，面向编程、逻辑推理与智能体任务，采用混合专家架构，每 token 仅激活 5010 亿参数中的 230 亿，主打计算效率。Reflection 称 Beam 在关键推理基准上可匹敌 GLM 5.2，但计算量少三到四倍，并接近规模更大的 Qwen3.8-Max。该模型基于 10500 块 Nvidia GB300 GPU 进行四周强化学习训练，将于本月晚些时候以 Apache 2.0 许可证发布权重。
正文节选
Reflection's Beam becomes the most capable open-weight model built outside China Key Points - AI startup Reflection is releasing Beam, its first open-weight model built for coding and reasoning, with a focus on compute efficiency over raw performance. - Beam activates just 23 billion of its 501 billion parameters per token and matches GLM 5.2 on key benchmarks while using three to four times less compute, according to Reflection. - The model was trained with reinforcement learning on 10,500 Nvid