StepFun releases natively multimodal Step 5 Preview, targeting coding and long-horizon agent tasks

StepFun released Step 5 Preview on September 20. The natively multimodal model takes text and vision inputs and is aimed at coding, software engineering, professional knowledge work and long-horizon agent tasks; at launch it includes 30 days of free Step Plan access.

On Artificial Analysis's intelligence index, Step 5 Preview briefly tied Kimi K3 for second among open-source models at 44 points, then fell to third after Xiaomi MiMo-V2.6 was released. It did not appear in OpenRouter's weekly top ten by call volume. The launch follows a January Series B+ of over $714M (RMB 5B) and a May pre-IPO round of nearly $2.5B, with Yin Qi as chairman.

Original sources (Chinese)