- StepFun opened API access to Step 5 Preview on September 20, a 600-billion-parameter sparse model that activates 27 billion parameters per token and reads up to 1 million tokens of context.
- Step 5 scores 44 on the Artificial Analysis Intelligence Index, level with Kimi K3 Max and roughly a seventh the cost of GPT-5.6 Sol, priced at $1 per million input tokens and $2.70 per million output tokens.
- Full open weights are due October 15, though the Hugging Face repository currently holds only a placeholder file.
StepFun's Step 5 matches Kimi K3 Max at a fraction of frontier pricing
Shanghai-based StepFun released Step 5 Preview on September 20 and opened its API the same day. The model uses a sparse mixture-of-experts design with 600 billion total parameters and 27 billion active per token, arranged in what the company describes as a 92-layer narrow-deep stack built for long-horizon agent work. Its context window reaches 1 million tokens, enough to hold a large codebase or a stack of long documents in a single session.
The pricing is where Step 5 makes its argument. At $1 per million input tokens and $2.70 per million output tokens, with a 95 percent discount on cached input, it undercuts the frontier by a wide margin while scoring 44 on the Artificial Analysis Intelligence Index. That number puts it level with Kimi K3 Max and roughly one seventh the price of OpenAI's GPT-5.6 Sol at a comparable rung of capability. StepFun is one of a cluster of well-funded Chinese labs, alongside the teams behind DeepSeek, Kimi, Qwen, and Zhipu, that has spent 2026 shipping capable models at prices that reset expectations for the rest of the market.
| Parameters | 600 billion total, 27 billion active per token (sparse mixture-of-experts) |
| Context window | 1 million tokens |
| Intelligence Index | 44 on the Artificial Analysis Intelligence Index |
| Pricing | $1 input, $2.70 output per million tokens, with a 95% cache discount |
| Open weights | Full release due October 15, 2026 |
Open weights on October 15 would put a frontier-class agent model in anyone's hands
The API is live now, and the more consequential date is October 15, when StepFun says it will publish the full weights. Releasing a model that matches a paid frontier tier on intelligence, at no license cost and runnable on a company's own hardware, keeps pressure on the closed labs where it hurts most, on price and on the reasons an enterprise pays a premium at all. The pattern has become familiar across China's open-weight releases this year, and it feeds directly into the open-source push Beijing has been backing as a lever against US model leaders.
| Measure | Step 5 Preview | Reference point |
|---|---|---|
| Intelligence Index | 44 | Level with Kimi K3 Max |
| Output price, per million tokens | $2.70 | About one seventh of GPT-5.6 Sol |
| Context window | 1 million tokens | Built for long-horizon agents |
| Availability | API live September 20 | Open weights due October 15 |
A caveat sits next to the launch. The weights are not out yet, and the Hugging Face repository listed for Step 5 currently ships only a .gitattributes file, so the independent benchmarking and red-teaming that decide a model's real standing cannot begin until the release lands. StepFun has made a strong claim on paper and priced it to move. Whether Step 5 holds up outside the company's own evaluations will be clear only once the weights are in the open and other researchers can run them for themselves.
Santage is committed to independent, transparent journalism. This article is produced in accordance with Santage's Editorial Standards and aims to provide accurate and timely information. Readers are encouraged to verify information independently.