阶跃星辰发布 Step 5 Preview：600B 总参 27B 激活的 MoE 模型，支持 1M 上下文

StepFun has released Step 5 Preview, its new flagship model for agentic work. The target workloads are software engineering, professional knowledge work, and finance. The main pitch is cost. StepFun team states the model delivers comparable intelligence at a substantially lower task cost. That is the ‘Pareto frontier’ framing in the launch title.

Is it deployable? Yes, as a hosted API and on the StepFun platform. Self-hosting waits for open weights. StepFun says open weights land on October 15, 2026. By simple arithmetic, 600B parameters need about 1.2 TB in BF16, before KV cache. Plan for multi-GPU server hardware once weights ship.

What StepFun Shipped

Step 5 Preview is a sparse Mixture-of-Experts (MoE) model. It holds about 600B total parameters and activates about 27B per token. That is roughly 4.5% of the weights per token.

The official model documentation lists these specs:

• Model ID: step-5-preview
• Context window: 1M tokens
• Input: text, images, and video
• Output: text
• Reasoning effort: low, medium, and high
• Streaming, tool calling, JSON Mode, JSON Schema, and prompt caching

On research tasks, StepFun team states the model coordinated 950 web fetches in a single agent action. StepFun team also documents a Claude Code integration through its Step Plan.

Architecture: Narrow and Deep

StepFun did not widen the network. It stacked 92 Transformer layers in a narrow-deep layout, according to Pandaily. The research team argues deeper stacks give longer paths for implicit multi-hop reasoning. This matters during long prefill, when agents search, run code, and read tool returns.

Training leans on on-policy, long-horizon reinforcement learning. StepFun cites bit-wise train and inference alignment across MoE routing. Other listed techniques include MTP-3 speculative decoding, FP8 MoE, and KV-cache offload. StepFun reports more than 3x end-to-end speedup for long-horizon RL.

Interactive Explainer

Benchmarks: Company-Reported vs Independent

Step 5 Preview ran at High effort, while rivals ran at Max.

StepFun reports these results, via RuntimeWire:

On coding, StepFun reports 67.7 on DeepSWE v1.1, 49.0 on StepCodeBench, and 80.5 on ProgramBench. GPT-6 Astra and Claude Opus 5 stay ahead on all 3. StepCodeBench is StepFun’s own benchmark.

StepFun also ran 2 agent experiments lasting 24 hours each. In the first, the model tuned an H100 kernel to 508 TFLOPS, against 493 for Claude Opus 5. In the second, it raised Qwen3-30B-A3B on AIME24 from 53.3% to 60% through automated post-training.

The independent check comes from Artificial Analysis. It scores Step 5 Preview at 44 on its Intelligence Index. The median for reasoning models in a similar price tier is 24. It measured output at 99.8 tokens per second on StepFun’s API.

Pricing

StepFun’s API list prices per 1M tokens:

Artificial Analysis puts the medians for comparable models at $1.88 input and $10.00 output. There is 1 catch. The model generated 160M output tokens on the index run, against a 92M median. Verbose reasoning eats part of the per-token savings.

Key Takeaways

• StepFun’s Step 5 Preview is a 600B-total, 27B-active MoE model.
• It offers a 1M-token context with text, image, and video input.
• API pricing is $1.00 input and $2.70 output per 1M tokens.
• Artificial Analysis scores it 44 on its Intelligence Index.
• Open weights are scheduled for October 15, 2026.

FAQ

1. What is Step 5 Preview? It is StepFun’s flagship MoE model for agentic coding, knowledge work, and finance.
2. Is Step 5 Preview open weight? Not yet. StepFun schedules open weights for October 15, 2026.
3. How large is the context window? 1M tokens, per StepFun’s documentation.
4. How much does it cost? $1.00 per 1M input tokens and $2.70 per 1M output tokens.