InternVL-U 4B
Shanghai AI Laboratory · 4B dense
Unified understanding, generation and editing at 4B
A unified 4B model that puts image understanding, generation and editing in one architecture, under Apache-2.0 and runnable on consumer GPUs. Generation 85 is its standout score, with perception 84 and documents 82 serviceable; video 65 and agents 55 show it is a content tool, not a business workhorse. Composite 75.8 puts it just inside the watchlist.
Last updated Oct 21, 2026Editorial scoring
Bottom line
Eight-dimension breakdown
editorial mapping · 0–100- InternVL-U 4B75.8
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 | 4B | Official model card |
| 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 |
|---|---|---|
| Product-image editing | 86 · Stable local redraw; text blurs easily | |
| Chinese poster generation | 84 · Usable layout; typos in long copy | |
| Chinese receipt OCR | 78 · Fine on clean receipts | |
| Mobile GUI control | 52 · Not viable |
Strengths and limitations
Strengths
- Understanding, generation and editing in one 4B model, deployable on a single consumer GPU.
- Apache-2.0 with no extra terms for commercial use or redistribution.
- Generation 85: usable image editing that is rare at this size.
Limitations
- Video 65 and agents 55 make it largely unusable for business automation.
- Documents 82: complex tables and handwritten receipts still need a larger specialist.
- The unified design means no single task is best-in-class; it cannot replace dedicated OCR or generation models.
Price and real cost
List price
—N/A
Open weights; runs on consumer GPUs. 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 InternVL-U 4B
Is InternVL-U 4B 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 InternVL-U 4B cost?
Open weights; runs on consumer GPUs. It ranks #47 with a composite of 75.8. Real bills depend on resolution, tiling, caching and volume discounts.
What is InternVL-U 4B best for?
Edge content-editing workbench; Product-image redraw and text replacement; Low-cost image QA demos。
Why does InternVL-U 4B score 75.8?
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 Ecosystem, cost & availability; its weakest is GUI agents & tool use.
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.