StepFun Releases Step 3.7 Flash, Focused on Agent Efficiency
阶跃星辰 Step 3.7 Flash 发布,聚焦智能体效率
StepFun released the open-source Step 3.7 Flash model with a 198B-parameter MoE architecture, about 11B active parameters, a 256K context window, and a 67.1 score on ClawEval-1.1.
Why it matters: HKR-H/K/R all pass: the release has a clear sparse-model hook, concrete context and benchmark numbers, and practitioner resonance around open agent efficiency. Official-post sourcing and no independent eval keep it in the 78–84 band.