StepFun and Together Compute will hold the Step 5 Preview Launch Night & AI Founders Meetup in San Francisco on October 22, featuring the first look at Step 5 Preview, live Q&A, and a panel on what to build as AI models improve, with Hongyu Zhou, Dan Fu, Mar Hershenson, Saoud Rizwan and Joey Wang on stage.
A week after its launch, Step 5 Preview reached No. 2 on OpenRouter's weekly leaderboard, with cumulative usage of 19.5T tokens, and has held the No. 2 spot on the daily leaderboard since October 10.
StepFun officially released Step 5 Preview weights, including BF16 and FP8 model weights, a technical blog, and open-source development tools. Designed for long-horizon agent tasks, the sparse MoE model has 600B total parameters, 27B activated parameters per token, a 1 million-token context window, and native text and vision input support. Weights are available on Hugging Face and ModelScope.
StepFun's Step 5 Preview model is now available on OpenRouter, offering flagship-tier intelligence for agentic and professional scenarios with significantly lower task costs. Free access starts today for one week across platforms including opencode, Cline, NousResearch, and Kilocode.
China has approved four more AI agent smartphones for network access, including three Honor Magic 9 series models and the STEPX Neo, the first LLM-native AI agent smartphone developed by StepFun subsidiary STEPX, following the Nubia NaviX Ultra’s approval in late August.
StepFun has released and open-sourced Step Code v0.1.0, a CLI-based AI coding tool under the MIT License, covering the full development workflow from code editing to testing and delivery. It scored 80.9% on Terminal-Bench 2.1 and 73.3% on Multi-Frame, a 150-task long-horizon benchmark, and supports one-command static site publishing via StepPage.
StepFun released Step 5 Preview, a new open-source flagship foundation model using a sparse MoE architecture (600B total parameters, 27B activated), natively supporting text and vision inputs with a 1M context length, targeting AI coding, software engineering, professional knowledge work, and finance. It scored 44 on the Artificial Analysis Intelligence Index, ranking among the top three open-source models globally.
StepFun released the StepAudio 3 series of voice models, including five models now available on its open platform, with several ranking first globally on Artificial Analysis benchmarks.