Step3-VL-10B
StepFun · 10B dense
A 10B starter base for image, text and video
A 10B vision-language model open-sourced by StepFun in June 2026, Apache-2.0 and 128K: perception 88, documents 86, charts 82 and a composite of 80.4. Its value is single-GPU deployment, fine-tunability and clean licensing, not the score; video 74, generation 55 and agents 72 are the boundaries of an entry tier.
Last updated Oct 21, 2026Editorial scoring
Bottom line
Eight-dimension breakdown
editorial mapping · 0–100- Step3-VL-10B80.4
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 |
|---|---|---|
| 参数规模 | 10B | Dense model |
| 许可证 | Apache-2.0 | Commercial use |
| 上下文 | 128K | Official spec |
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 | 86 · Fine on clean receipts | |
| Chart data extraction | 82 · Stable on simple charts | |
| 30-min video QA | 72 · Works on sparse frames | |
| Mobile GUI control | 68 · Not GUI-specific | |
| Chinese cultural context | 84 · General context |
Strengths and limitations
Strengths
- 10B plus Apache-2.0: the lowest barrier for single-GPU deployment and coursework.
- Perception 88 and documents 86 are solid for an entry tier.
- Grade A: weights, evaluation scripts and model card are public.
Limitations
- Video 74: sparse-frame understanding only.
- Generation 55: no usable image generation or editing.
- Agents 72 and ecosystem 88: complex orchestration and hosted services must come from elsewhere.
Price and real cost
List price
—N/A
Open weights; runs on one GPU. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥0.0 (assuming input at 25% of output price).
Use your volumeAbout Step3-VL-10B
Is Step3-VL-10B open source?
Licence: Open weights (Apache-2.0). Weights are available, but check the exact terms and filing requirements before commercial use.
How much does Step3-VL-10B cost?
Open weights; runs on one GPU. It ranks #35 with a composite of 80.4. Real bills depend on resolution, tiling, caching and volume discounts.
What is Step3-VL-10B best for?
Teaching and research prototypes; Edge image-text experiments; A starting point for domain fine-tuning。
Why does Step3-VL-10B score 80.4?
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 Generation & editing.
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.