Step 5 Preview
Step 5 Preview is a preview-stage chat model from Chinese AI developer StepFun, available through hosted inference APIs and tracked here for throughput and pricing comparison.
API Pricing
Cheapest input on OpenRouter · cheapest output on OpenRouter| Provider | Input / 1M | Output / 1M | Cached / 1M |
|---|---|---|---|
| $1.00 | $2.70 | $0.100 | |
| $1.00 | $2.70 | $0.100 | |
| $1.00 | $2.70 | $0.100 | |
| $1.00 | $2.70 | $0.100 |
Prices updated daily. Last check: Nov 5, 2026
Step 5 Preview pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Benchmarks measured Nov 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- StepFun
- Modalities
- Text
Capabilities
- Open Source
- No
Strengths & Limitations
Strengths
- Measured output throughput of approximately 89.8 tokens per second in third-party serving benchmarks
- Preview access allows developers to test and integrate before a general-availability release
- Part of StepFun's Step model series, giving a migration path within the same family and API surface
- Available through hosted inference APIs, so no local GPU provisioning is required to evaluate
- Throughput and pricing are tracked across providers on this page for direct comparison
Limitations
- Preview release — behavior, pricing, and availability may change without notice before general availability
- Time to first token of roughly 1.9 seconds is slow for latency-sensitive interactive chat
- Context window is not confirmed in our database, which complicates planning for long-document workloads
- Limited published benchmark coverage compared with widely evaluated Western models
- Documentation and community support are primarily oriented toward the Chinese developer ecosystem
Key Features
About Step 5 Preview
Common Use Cases
Step 5 Preview suits teams that want early access to a StepFun chat model for evaluation — prototyping assistants, drafting and summarization pipelines, and internal tooling where the roughly 90 tokens/sec generation rate is adequate and the near two-second time to first token is acceptable. The latency profile favors asynchronous and batch workloads: document processing queues, offline content generation, scheduled report drafting, and agent steps where a human is not watching the cursor. It is a reasonable candidate for organizations already working within the Chinese AI ecosystem or benchmarking Chinese-developed models against alternatives. Because it is a preview build, it is better suited to experimentation and capability testing than to production systems that require stable, long-lived model behavior.
Frequently Asked Questions
How much does Step 5 Preview cost?
Pricing varies by provider and by pricing type — input versus output tokens, cached input, and any batch or committed-use discounts each differ. Rates also change over time. See the pricing table on this page for the current per-provider figures.
What is Step 5 Preview best used for?
It fits general chat and text generation tasks where a sub-second first-token response is not required — summarization, drafting, batch content processing, and asynchronous agent steps. Its measured ~90 tokens/sec output rate makes it workable for longer generations, while the ~1.9 second time to first token makes it less comfortable for snappy interactive chat.
What does the 'Preview' in the name mean?
Preview indicates a pre-general-availability release from StepFun. Preview models are published so developers can evaluate and integrate early, but their behavior, pricing, and availability can change before — or instead of — a stable release. Avoid depending on exact output behavior in production systems.
How fast is Step 5 Preview?
Third-party measurements from Artificial Analysis report approximately 89.8 output tokens per second with a time to first token around 1,882 ms. Real-world numbers depend on the provider, prompt size, region, and current load.
What is the context window of Step 5 Preview?
We do not currently have a confirmed context window for Step 5 Preview in our database. Check StepFun's official API documentation for the authoritative limit before designing long-context workloads around it.
Who makes Step 5 Preview?
It is developed by StepFun, a Chinese AI company behind the Step series of foundation models. The model is accessed through hosted inference APIs listed on this page.