VideoLLaMA 3
Open 7B foundation model for joint image and video understanding.
About
VideoLLaMA 3 is the third generation of Alibaba DAMO Academy's VideoLLaMA series, released in January 2025 by the Singapore NLP group. Built on SigLIP for vision and Qwen2.5 for language, it handles both images and video in a single model, including temporal grounding and fine-grained object understanding across frames.
The code is Apache-2.0 and the authors claim state-of-the-art results on mainstream video understanding benchmarks in their technical report. The model weights inherit Qwen licensing terms and the hosted demo is explicitly a non-commercial research preview, which limits how companies can deploy it. There is no official hosted API; developers self-host the 7B checkpoint from Hugging Face.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Authors claim state-of-the-art performance on mainstream video understanding benchmarks including VideoMME (technical report, arXiv 2501); 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 — 6/10
Scored 6/10 (metrics, metrics-v1). Evidence: DAMO-NLP-SG official GitHub, Hugging Face weights, technical report. Strengths: permissive code license, strong academic lineage, unified image+video model. Docked for: (1) weights inherit Qwen license terms and the demo is research-only non-commercial, creating deployment friction; (2) no official API and modest community adoption versus Qwen-VL siblings. License: Apache-2.0 (code); weights subject to Qwen model license.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · Alibaba DAMO Academy (Singapore NLP), Jan 2025, SigLIP + Qwen2.5
- · Unified image+video 7B model; no official hosted API