Step 5 Preview by JieYue: A 600B Sparse MoE Model with Merely 27B Active Parameters, Redefining Cost-Effectiveness in Open-Source Flagship Models
5 day ago / Read about 0 minute
Author:小编   

JieYue has unveiled its latest flagship base model, the Step 5 Preview, which is primarily designed to tackle real-world Agentic tasks. This innovative model leverages a sparse Mixture of Experts (MoE) architecture, boasting a staggering 600 billion parameters in total, yet only activating 27 billion of them during operation. It accommodates an extensive context window of up to 1 million tokens and seamlessly processes both textual and visual inputs. The Step 5 Preview shines in various scenarios, including AI programming, software engineering, professional knowledge-intensive work, and finance. It has secured a remarkable score of 44 on the globally acclaimed Artificial Analysis Intelligence Index, placing it among the top three open-source models worldwide. Moreover, the cost per task for this model is a mere fraction, specifically one-eighth, of that incurred by Claude Opus 5. JieYue is set to release the model weights on October 15, marking a significant milestone in the field.

  • C114 Communication Network
  • Communication Home