Ovis-U1
Unified 3B model for vision understanding, image generation, and editing.
About
Ovis-U1 is a 3-billion-parameter unified multimodal model from AIDC-AI, released in June 2025 with a technical report (arXiv:2506.23044). Unlike models that only understand images, Ovis-U1 folds understanding, text-to-image generation, and image editing into a single framework, built on the earlier Ovis visual-embedding architecture and FLUX-style generation.
The project is Apache-2.0, which is unusually permissive for a unified model of this kind, and it has picked up traction in the open community, including ports for Huawei's Ascend NPUs. It is known for strong document and OCR-adjacent performance for its size, though the authors include a compliance disclaimer about training-data provenance. There is no official hosted API; the model is self-hosted from Hugging Face.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Authors claim competitive results on mainstream multimodal benchmarks in the technical report (arXiv:2506.23044); no single headline figure independently extracted here
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified, metrics-v1). Evidence: AIDC-AI official GitHub (aidc-ai/ovis-u1), HF AIDC-AI/Ovis-U1-3B, technical report arXiv:2506.23044, Apache-2.0 license in repo. Strengths: rare permissive unified understand-plus-generate model, active community ports. Docked for: (1) training-data compliance disclaimer from the authors themselves; (2) smaller team footprint and no hosted option. License: Apache-2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · AIDC-AI (Alibaba), June 2025; technical report arXiv:2506.23044
- · 3B unified understanding + generation + editing model
- · Authors include a training-data compliance disclaimer