StepFun launched Step 5 Preview on September 18, 2026, a 600-billion-parameter reasoning model that outperforms most competitors on intelligence benchmarks while charging a fraction of the typical price. The model scored 44 on the Artificial Analysis Intelligence Index, nearly double the median score of 25 for models in its price tier.
What Step 5 Preview Costs and How It Compares
StepFun charges $1.00 per million input tokens and $2.70 per million output tokens through its API. For comparison, the median pricing among comparable reasoning models sits at $1.88 for input and $10.00 for output. A blended rate using a 7:2:1 cache hit to input to output ratio comes to $0.51 per million tokens.
Artificial Analysis spent $918.34 to evaluate Step 5 Preview on its Intelligence Index. The model generated 160 million output tokens during testing, well above the median of 90 million for its peer group. That verbosity is a direct result of the extended thinking process the model uses to reason through problems before responding.
How the Intelligence Index Measures Performance
The Artificial Analysis Intelligence Index is a composite benchmark that tests models across reasoning, knowledge, mathematics, and coding. It includes evaluations for agentic knowledge work, real-world task completion, SaaS workflows, coding and terminal use, professional document reasoning, physics reasoning, legal agentic work, quantitative analysis on spreadsheets, Kubernetes incident root-cause analysis, visual reasoning, and medical long-context reasoning. The index also measures hallucination rate and long-context reasoning ability.
Step 5 Preview was classified as a proprietary model, meaning its weights are not publicly available. As such, it was compared against both proprietary and open-weights models within the same price range, using a blended 3:1 input-to-output price ratio.
Technical Specs and Capabilities
The model supports text and image input and generates text output, making it multimodal. Its context window spans 1 million tokens, roughly equivalent to 1,500 pages of A4 text at 12-point Arial. With 600 billion parameters, Step 5 Preview falls into the large model category.
Step 5 Preview uses extended thinking, a form of chain-of-thought reasoning where the model works through complex problems step by step before producing a final answer. This approach is similar to techniques used by OpenAI's o1 and o3 models and Anthropic's Claude models with thinking enabled.
What This Means for Developers
The price-to-performance gap is the most notable aspect of Step 5 Preview. Output token costs are 73 percent lower than the median for comparable reasoning models, which can significantly reduce inference expenses for teams building applications that require heavy reasoning. The 1 million token context window gives it an edge for document analysis, codebase review, and long-running conversations.
The tradeoff is verbosity. Step 5 Preview generates roughly 78 percent more output tokens than the median reasoning model during evaluation. That means longer responses and potentially higher total costs per task if prompts generate many turns. Teams building applications that need short, direct answers may need to add prompt constraints or post-processing to manage token consumption.
Access is limited. Step 5 Preview is available through a single API provider, which creates a dependency risk. There is no option to self-host or fine-tune the model since the weights are proprietary. Teams that need to keep data on-premises or modify model behavior will need to consider alternatives.
Where Step 5 Preview Fits in the Market
The reasoning model market has expanded rapidly since OpenAI introduced o1 in late 2024. Google, Anthropic, DeepSeek, and now StepFun all offer models with extended thinking capabilities. Step 5 Preview distinguishes itself through cost efficiency rather than raw benchmark supremacy. For teams that have been paying premium prices for reasoning quality, it offers a viable path to lower costs without sacrificing performance on complex tasks.
The model's strong showing on agentic business operations, quantitative analysis, and Kubernetes root-cause analysis suggests StepFun is targeting enterprise and infrastructure use cases. With its combination of multimodal input, long context, and aggressive pricing, Step 5 Preview gives developers a reason to look beyond the dominant model providers when the task calls for deep reasoning.