StepFun· @StepFun_ai · X·· 2026-05-29AI 评分64
AI 导读
SGLang宣布其推理框架现已对阶跃星辰（StepFun AI）的新模型Step 3.7 Flash提供Day-0支持。该模型是一个专为高吞吐量智能体工作负载设计的198B稀疏MoE视觉语言模型，具备原生多模态感知能力。它在SimpleVQA评测中以79.2分排名第一，V*得分为95.3。在智能体工作流方面，它在ClawEval-1.1上以67.1分领先，展现出可靠的长期工具编排能力；在软件工程任务中，于SWE-Bench PRO上以56.3分位列第二。模型支持256K上下文，并提供3种推理级别以平衡速度、成本与深度。
正文 · 原文
Day-0 SGLang support. Thanks @lmsysorg for moving so fast.
🎉 Meet Step 3.7 Flash from @StepFun_ai, a 198B sparse MoE vision-language model built for high-throughput agentic workloads. Day-0 support is now live in SGLang! > Native multimodal perception: #1 on SimpleVQA 79.2, V* at 95.3, turns UIs & charts into code > Workflow integrity: leads ClawEval-1.1 at 67.1, reliable long-horizon tool orchestration > Live engineering: 2nd on SWE-Bench PRO 56.3, traces repos, isolates bugs, ships patches > 256K context + 3 reasoning levels to balance speed, cost & depth Run it now with SGLang!在 X 查看被引用的帖子
来源：StepFun · x.com