StepFun Launches Step 5 Preview, a 600-Billion-Parameter AI Model with 1 Million Token Context
Chinese AI company StepFun launched the Step 5 Preview API on September 20, offering a 600-billion-parameter sparse mixture-of-experts model with a 1 million token context window. The model is priced at 1 dollar per million input tokens and is available now, with open weights scheduled for release on October 15.
StepFun launched the Step 5 Preview API on September 20, 2026. The model uses a sparse mixture-of-experts architecture with approximately 600 billion total parameters, of which 27 billion are active per token. It supports a 1 million token context window and accepts text and image inputs, with text-only output.
The model is capable of reasoning with extended thinking. StepFun is positioning it as a flagship for agentic tasks, coding, and professional knowledge work.
Independent evaluations by Artificial Analysis assigned the model an Intelligence Index score of 44, ranking it 24th out of 200 models. On Terminal-Bench 4.0, it achieved 33.3 percent, outperforming several peers. Early developer reports have noted challenges with tool-calling accuracy and potential output truncation under heavy load.
The model is priced at 1 dollar per million input tokens and 2.70 dollars per million output tokens, with a 95 percent discount applied to cached tokens. It is available through the StepFun API and Studio.
The Hugging Face repository for the model currently serves as a placeholder. StepFun has scheduled the final open-weights release for October 15, 2026.
The Step 5 Preview launch comes as the broader AI industry shifts toward agentic workflows, where models plan and execute multi-step tasks rather than simply generating text. Other major releases in September 2026 include OpenAI's GPT-6 Astra, which was rated "Critical" for cybersecurity risk after demonstrating the ability to independently discover and exploit vulnerabilities in hardened systems.
