Step5 Preview
A preview release of StepFun's flagship large language model, built on a Mixture-of-Experts (MoE) architecture with 600 billion total parameters and 27 billion activated parameters. It supports visual understanding and context windows of up to one million tokens.
Timeline (last 90 days)
StepFun Step5 Preview is a 600-billion-parameter MoE model with only 27 billion activated parameters, supporting visual data processing and a million-level token context window.
StepFun's official charts claim that Step5 Preview is benchmarked against GLM-5.3 and comparable Qwen models, with per-task costs reduced by approximately 65%.
StepFun has stated that it will release the open-source weights for Step5 on October 15.
All Facts (2)
StepFun's official charts claim that Step5 Preview is benchmarked against GLM-5.3 and comparable Qwen models, with per-task costs reduced by approximately 65%.
50%UnverifiedStepFun Step5 Preview is a 600-billion-parameter MoE model with only 27 billion activated parameters, supporting visual data processing and a million-level token context window.
50%