StepFun Releases Step 5 Preview: 600B MoE Open Weights
StepFun has launched Step 5 Preview with open weights on Hugging Face and ModelScope, featuring a 600B MoE architecture, 27B active parameters, and a 1M token context window.
Stock photo for illustration only, not from the actual event
- StepFun has released Step 5 Preview as open weights on Hugging Face and ModelScope.
- The sparse Mixture-of-Experts model features a total of 600 billion parameters.
- It utilizes 27 billion active parameters per token with up to 1 million token context.
- Engineered specifically to handle long-horizon agentic workflows for developers.
The artificial intelligence community has a new tool to explore as StepFun officially rolls out Step 5 Preview. The model is now available as open weights, with both FP8 and BF16 checkpoints ready for developers to access directly on Hugging Face and ModelScope.
Step 5 Preview is built as a sparse Mixture-of-Experts (MoE) model, carefully structured by the StepFun team to handle demanding long-horizon agentic tasks that require sustained reasoning and complex multi-step execution over extended periods.
Stock photo for illustration only, not from the actual event
Under the hood, the model boasts an impressive scale with a grand total of 600 billion parameters, while operating efficiently with 27 billion active parameters per token. Furthermore, it supports an extensive context window of up to 1 million tokens.
Releasing open-weight models with massive parameter counts and 1-million-token context windows underscores the ongoing industry push toward advanced agentic automation. This capability allows developers to process vast amounts of data in a single pass, overcoming previous memory and reasoning limits.
Developers interested in exploring advanced agent workflows and evaluating the model's capabilities can now retrieve the checkpoints directly from the designated public repositories provided by StepFun.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment