StepFun shows Step 5 Preview inference performance on GB300
What happened
On October 15, StepFun showed off Step 5 Preview running on NVIDIA's GB300, saying inference performance is strong and inviting developers to build with it. The NVIDIA post StepFun shared calls Step 5 Preview its largest open-weight release: a 600B MoE architecture that can handle agent tasks lasting up to 24 hours. On GB300 with NVFP4, throughput reaches nearly 8K tokens/sec/GPU. Coverage so far centers on model size, agent task handling and GB300 inference numbers.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
- AI HOT · ModelsStepFun 展示 Step 5 Preview 在 NVIDIA GB300 上的推理性能
StepFun 表示 Step 5 Preview 在 NVIDIA GB300 上推理表现出色，并期待开发者用它构建应用。其转发的 NVIDIA 内容称，该模型是 StepFun 最大的开源权重发布，采用 600B MoE 架构，可处理最长 24 小时的 AI 智能体任务，在 GB300 上使用 NVFP4 达到近 8K tokens/sec/GPU。
Heat over time
Not enough continuous observations to draw a trend yet.