阶跃星辰把 Step-5-Preview 开源了,600B 参数只用激活 27B,输入输出都能吃 100 万上下文,长文档场景可以试试。
#Step-5-Preview
Stability across repeated agent runs is the standout trait here, which matters for anyone running long agent loops where flaky outputs compound. The model's rule-exploiting behaviour in the rider test suggests it is well suited to optimisation-heavy tasks under explicit constraints. Developers building agentic coding workflows should compare it against current SOTA models, while front-end and 3D work still call for alternatives.
The debugging loop matters more than the demo outputs: the model identifies failures, patches code and re-tests without prompt engineering, which is the practical bar for agentic coding use. Developers evaluating Chinese frontier models can weigh Step-5-Preview against similarly priced alternatives on the cost-intelligence curve. Financial professionals should note its reportedly strong domain performance in finance workflows.