Stepfun Open-Sources Step 3.7 Flash LLM Optimized for Agent Era
Stepfun (Jieyue Xingchen), a Chinese AI company, has officially released and open-sourced Step 3.7 Flash, a next-generation language model specifically optimized for the agent production era. The model is designed around agent workflows, coding, search, and multimodal capabilities. Built on a sparse MoE architecture, Step 3.7 Flash features 196B total parameters plus 1.8B ViT, with 11B activated parameters per inference. It achieves generation speeds of up to 400 tokens per s
Chinese AI firm Stepfun has open-sourced its latest language model, Step 3.7 Flash, engineered for the growing field of AI agents. This model boasts a sparse MoE architecture with a large total parameter count but a smaller active parameter set per inference, enabling rapid generation speeds of up to 400 tokens per second. It is designed to handle complex agent workflows, including native multimodal understanding of visual elements like UIs and images, alongside robust web and visual search capabilities. Step 3.7 Flash also excels at reliable tool calling and orchestration, seamlessly integrating with various agent frameworks and protocols to execute tasks across different applications and systems.
This release provides developers with a powerful, open-source tool optimized for multimodal understanding and reliable tool integration, accelerating the development of sophisticated AI agent applications.
📌 Kaynak
Bu özet Pandaily kaynağından otomatik derlenmiştir. Tamamı için orijinal habere gidin.
Orijinal haberi oku →