Step 5 Preview from Jieyue opens its large model weights, supporting a context window of up to one million tokens.
Step5Preview uses a sparse MoE architecture with a total parameter count of 600B and only 27B activations per token. It natively supports text and visual inputs and is designed for long-horizon agent tasks. Model weights are available on Hugging Face and ModelScope and can be deployed using vLLM, SGLang, or Transformers.
Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open the original article on ITHome