Step 5 Preview

StepFun 600B MoE (27B active) multimodal flagship preview on NVIDIA NIM in NVFP4 — 1M context, 40 RPM.

Updated 11/22/20261 code examples

About the Model

为什么要选择 Step 5 Preview?

Step 5 Preview 是阶跃星辰(StepFun)推出的新一代旗舰开源基础模型预览版。采用 600B 总参数、27B 动态激活的稀疏 MoE 架构,支持 100 万(1M)Token 原生超长上下文,并经 NVIDIA NVFP4 深度量化加速,专为长流程代码构建、复杂文档解析与自主 Agent 规划设计。

核心优势与技术突破

  • 600B 旗舰 MoE 架构:27B 动态激活兼顾顶级推理规划性能与高并发吞吐。
  • 100 万 Token 上下文与原生多模态:支持单次灌入整个中大型代码仓库或数百页技术规格书与设计图纸。
  • NVFP4 原生硬件加速:在 NVIDIA DGX Cloud 上以 NVFP4 精度部署,推理延迟大幅降低,显存开销极大收敛。

How to Access for Free (via NVIDIA NIM)

Why Choose Step 5 Preview on NVIDIA NIM?

Experience StepFun's 600B MoE flagship model (stepfun-ai/step-5-preview) deployed on NVIDIA NIM with NVFP4 precision. Featuring 27B dynamic active parameters, 1M context, and native multimodal vision, it excels at complex code refactoring and multi-step agent reasoning.

Try it in your browser

Pick a model, paste your own free key, and run. Your key is sent once to call the provider and never stored on our servers.

Pick a model, paste your own free key, and run. Your key is sent once to call the provider and never stored on our servers.

Code Examples

curl
curl https://integrate.api.nvidia.com/v1/chat/completions \
  -H "Authorization: Bearer $NVIDIA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "stepfun-ai/step-5-preview",
    "messages": [{"role": "user", "content": "设计一个具备背压机制的高吞吐异步事件调度器。"}]
  }'

Community Feedback

Comments are tied to your GitHub account — sign in to join the discussion.