Step-3
StepFun · Undisclosed (MoE)
A practical edge-to-cloud multimodal stack
Step-3 is balanced across vision, speech and video, with edge variants for phones and cars. No runaway strength, but good price and latency.
Last updated Oct 18, 2026Editorial scoring
Bottom line
Eight-dimension breakdown
editorial mapping · 0–100- Step-384.7
Weights: perception 20 / documents 18 / charts 14 / video 14 / agents 14 / generation 6 / Chinese 6 / ecosystem 8. Definitions and sources on the methodology page.
Public benchmarks
as reported, not converted| Benchmark / metric | Value | Note |
|---|---|---|
| 上下文 | 256K | Official spec |
| 模态 | 4 | Text/image/audio/video |
Public scores come from different versions and prompts and are not interchangeable with our composite; they are one evidence input.
Scenario mapping
five stress tests, one scale| Scenario | Score | Observation |
|---|---|---|
| Chinese receipt OCR | 85 · Standard documents | |
| Chart data extraction | 83 · Standard | |
| 30-min video QA | 86 · Strong on short video, average long | |
| Mobile GUI control | 85 · Fast on-device response | |
| Chinese cultural context | 84 · General |
Strengths and limitations
Strengths
- Mature edge variants with low latency.
- Balanced speech and vision.
- Moderate price for mass deployment.
Limitations
- Fewer public evals and docs.
- Newer ecosystem and tooling.
- No single standout dimension.
Price and real cost
List price
¥5/M tokens
≈ ¥5 / 1M output tokens. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥3.0 (assuming input at 25% of output price).
Use your volumeAbout Step-3
Is Step-3 open source?
Licence: Closed commercial. No public weights; access is via vendor API or an on-prem licence.
How much does Step-3 cost?
≈ ¥5 / 1M output tokens. It ranks #17 with a composite of 84.7. Real bills depend on resolution, tiling, caching and volume discounts.
What is Step-3 best for?
Edge-cloud products; In-car multimodal interaction; Light speech-plus-vision apps。
Why does Step-3 score 84.7?
The composite weights eight dimensions: perception 20%, documents 18%, charts 14%, video 14%, agents 14%, generation 6%, Chinese 6%, ecosystem 8%. Its strongest dimension is Visual perception & grounding; its weakest is Ecosystem, cost & availability.
Personal ranking · Not an official score
Every ranking, score and comparison on this page is an independent personal ranking, not an official score, and represents no vendor, institution or regulator. Scores are a subjective mapping of public evidence under our own weights — change the weights and the order changes. Disagree, or spotted a factual error? Contact the site owner: jizhi2024@foxmail.com — we check each report and log corrections in the changelog.