Loading page
StepFun
Step 3.7 Flash
Step 3.7 Flash is a large language model from StepFun, positioned in the Step family's "Flash" line, with measured throughput of roughly 127 output tokens per second.
Input from
$0.200 / 1M tokens
across 4 providers
API Pricing
| Provider | Input / 1M | Output / 1M | Cached / 1M |
|---|---|---|---|
| $0.200 | $1.15 | $0.040 | |
| $0.200 | $1.15 | $0.040 | |
| $0.200 | $1.15 | $0.040 | |
| $0.200 | $1.15 | $0.040 |
Prices updated daily. Last check: Sep 20, 2026
Step 3.7 Flash pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Intelligence
19.5 / 100
Coding
39.6 / 100
Output Speed
179 t/s
Latency (TTFT)
1.4s
Reasoning & Knowledge
- GPQA Diamond80.9%
- Humanity's Last Exam21.4%
Coding
- SciCode43.9%
Agentic & Tool Use
- Terminal-Bench Hard35.6%
- Terminal-Bench v2.139.3%
- τ²-bench98.5%
- τ-bench Banking12.0%
Instruction & Long Context
- IFBench67.3%
- Long-Context Reasoning73.7%
Benchmarks measured Sep 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- StepFun
- Modalities
- Text
Capabilities
- Tool Calling
- No
- Open Source
- No
Strengths & Limitations
Strengths
- Measured output throughput of about 126.7 tokens per second, suitable for streaming chat interfaces
- Positioned in the Step family's Flash tier, which targets faster generation than heavier variants in the series
- Independently benchmarked for serving speed by Artificial Analysis rather than relying only on vendor claims
- Offers an alternative to the more widely deployed US-origin model families for teams evaluating diverse suppliers
- StepFun develops models with attention to Chinese-language use, making the Step series worth evaluating for bilingual workloads
- Throughput profile fits batch text generation and summarization pipelines where many requests run in parallel
Limitations
- Measured time to first token of roughly 1,743 ms is a consideration for latency-critical, single-shot interactions
- We do not currently track a confirmed context window for this model — verify limits with the serving provider
- Reasoning, coding, and knowledge benchmark scores are not available in our dataset, making capability comparisons difficult
- Provider availability for StepFun models is narrower than for the largest model families, which limits routing and failover options
- Flash-tier positioning implies capability trade-offs relative to larger variants in the Step series
Key Features
•Text generation with streaming output
•Measured throughput of approximately 126.7 output tokens per second
•Time to first token measured at approximately 1,743 ms
•Flash-tier variant within StepFun's Step 3.7 generation
•Independent serving benchmarks from Artificial Analysis
•Available through API endpoints from providers listed in the pricing table
•Candidate model for Chinese and bilingual text workloads
About Step 3.7 Flash
Step 3.7 Flash is a language model developed by StepFun, a Chinese AI company that develops the Step model series. The "Flash" designation within the Step lineup indicates a variant oriented toward faster generation rather than a maximum-capability configuration, making it a candidate for workloads where response latency and token throughput matter alongside output quality.
Independent measurements collected by Artificial Analysis put Step 3.7 Flash at approximately 126.7 output tokens per second, with a time to first token of roughly 1,743 milliseconds. The throughput figure places it in a range suited to interactive applications where text streams back to a user, while the time to first token is a factor to weigh for latency-sensitive front-end experiences. Beyond these serving metrics, computeprices.com does not currently track a confirmed context window, modality list, or reasoning benchmark set for this model, so those attributes should be verified against StepFun's own documentation or the serving provider's API reference before committing to a deployment.
In practice, models in a "Flash"-style tier are commonly used for high-volume text generation, chat assistants, and summarization pipelines where per-request speed and cost efficiency are weighted more heavily than peak reasoning depth. Because provider coverage for StepFun models is narrower than for models from the largest Western labs, availability and pricing vary considerably — the pricing table on this page shows which providers currently serve Step 3.7 Flash and at what rates.
Common Use Cases
Step 3.7 Flash fits workloads where generation speed and per-token efficiency matter more than maximum reasoning depth: chat assistants that stream responses to users, content drafting and rewriting, document and conversation summarization, data extraction from unstructured text, and high-volume classification or tagging pipelines. Its measured throughput of roughly 127 output tokens per second supports interactive experiences, while the time to first token of about 1.7 seconds suggests it is a better fit for streaming interfaces — where users see partial output quickly — than for ultra-low-latency single-token responses such as autocomplete. Teams evaluating Chinese-language or bilingual applications may want to include it in a benchmark set alongside models from other families, since StepFun develops primarily for that market. For tasks that demand long multi-step reasoning or verified long-context handling, confirm the model's context window and reasoning benchmarks with the provider before selecting it.
Frequently Asked Questions
How much does Step 3.7 Flash cost?
Pricing for Step 3.7 Flash varies by provider and by pricing type — input tokens, output tokens, and any cached or batch rates are typically billed separately, and different hosts set different rates. See the pricing table on this page for the current per-provider figures.
What is Step 3.7 Flash best used for?
It suits high-volume text workloads that benefit from fast generation: streaming chat assistants, summarization, content drafting, text extraction, and classification pipelines. The Flash tier positioning and measured throughput of about 127 output tokens per second point toward throughput-oriented rather than maximum-reasoning use.
How fast is Step 3.7 Flash?
Artificial Analysis measured approximately 126.7 output tokens per second with a time to first token of roughly 1,743 milliseconds. The throughput figure is well suited to streaming responses; the time to first token is worth testing against your own latency requirements.
What context window does Step 3.7 Flash support?
We do not currently track a confirmed context window for Step 3.7 Flash. Check StepFun's documentation or your serving provider's API reference for the maximum input length, as hosted deployments sometimes cap context below the model's native limit.
How does Step 3.7 Flash compare to other models in the Step family?
The Flash designation indicates a speed-oriented variant within the Step 3.7 generation, so it is expected to generate faster than heavier siblings while trading off some capability. Because we do not have comparable benchmark scores across the family in our dataset, running your own evaluation on representative prompts is the most reliable way to compare them.
Who develops Step 3.7 Flash?
Step 3.7 Flash is developed by StepFun, a Chinese AI company behind the Step series of large language models.