AITOPAI热报
ModelsOfficial72°

StepFun 发布 Step 5 Preview：约 600B MoE、1M 上下文，10 月 15 日开源权重

StepFun Step 5 Preview: ~600B MoE Long-Horizon Agent; Open Weights Due Oct 15

Primary sourcepandaily@[email protected] (Pandaily)
Blog/Media
Read the original
Why we picked it

StepFun 这个 Step 5 Preview 有 1M 上下文，主打长程智能体任务，权重月底开源，可以等开源后自己跑一跑。

StepFun 于 2026 年 9 月 20 日发布 Step 5 Preview，采用约 600B 参数的稀疏 MoE 架构，激活参数约 27B。

Body

模型支持 1M token 上下文，面向需要长程规划的智能体场景。API 已上线，BF16 格式的开源权重计划于 10 月 15 日放出。

Original · pandaily

StepFun Step 5 Preview: ~600B MoE Long-Horizon Agent; Open Weights Due Oct 15

StepFun launched Step 5 Preview on September 20, 2026: a ~600B sparse MoE (~27B active) with 1M context for long-horizon agents, API live now and BF16 open weights planned for October 15.

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison and AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include StepFun, Step 5, MoE, 开源模型.

Primary source verification: published directly by pandaily with first-party authority.

#StepFun#Step 5#MoE#开源模型#智能体

Keep reading

Recent stories with the same tag or section
ModelsUS vs ChinaSelected23:57
Together AI Triples StepFun Step 5 Serving Throughput

The interesting part is the method: AI agents drove the search across routing, scheduling and kernel-level changes, with a strict benchmark-before-merge loop — a reproducible pattern for anyone running inference infrastructure. Teams serving MoE models can apply the same TTFT- and TPS-targeted workflow. StepFun is a Chinese lab, so its models being served at scale on a US stack also signals deepening cross-border infrastructure ties.

#StepFun#Together AI#open-source models#inference
Official accountTogether AI@togethercomputeSource
ProductsBuildersSelected00:01
StepFun Showcases Step 5 Preview and Step Code at AI Engineer Shanghai 2026

The pairing of Step 5 Preview with an open-source Agent Harness signals that StepFun is treating agent reliability over long task horizons as a product-level problem, not just a model problem. Developers building agentic workflows can examine Step Code as an open alternative to proprietary agent frameworks from labs like OpenAI or Anthropic. The event also underscores StepFun's push to build a developer community around its stack.

#StepFun#Step 5 Preview#Step Code#agents
Official account阶跃星辰 Stepfun@Stepfun_AISource