VideoLLaMA3
Frontier multimodal foundation model for image and video understanding, built on SigLIP and Qwen2.5.
About
VideoLLaMA3 is DAMO Academy's video-understanding foundation model, built on SigLIP for vision and Qwen2.5 for language, with design lessons borrowed from LLaVA-OneVision, InternVL2 and Qwen2-VL. The January 2025 technical report positions it as a general image-and-video model with spatial-temporal modeling, continuing the Video-LLaMA line that started in 2023.
The code is Apache 2.0, but the weights are released as a research preview for non-commercial use, so commercial teams should read the license carefully. VideoLLaMA3-7B is on Hugging Face for self-hosting. Published evaluations are author-reported only, with no independent benchmark numbers to cite, and the fast-moving open video-model field has since produced strong alternatives. It is still a solid reference implementation for video-language modeling, especially for researchers already in the DAMO and Qwen ecosystem.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Author-reported evaluations in the technical report (arXiv 2501.13106); specific numbers not independently verified
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 (TIER metrics, metrics-v1). Evidence: arXiv 2501.13106 (Jan 2025) from DAMO-NLP-SG; GitHub DAMO-NLP-SG/VideoLLaMA3; HF DAMO-NLP-SG/VideoLLaMA3-7B. Strengths: frontier-positioned open video model, builds on proven SigLIP+Qwen2.5 stack, documented lineage from LLaVA-OneVision/InternVL2. Docked for: (1) evaluations are author-reported only, no independent numbers; (2) weights are non-commercial research preview, not truly open; (3) crowded field with newer alternatives. License: Apache 2.0 (code); research-preview non-commercial terms (weights).
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 6.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · Technical report arXiv 2501.13106 (2025-01-22) from DAMO-NLP-SG; SigLIP + Qwen2.5
- · GitHub DAMO-NLP-SG/VideoLLaMA3 verified live (curl 200, Oct 2026); HF DAMO-NLP-SG/VideoLLaMA3-7B
- · Code Apache 2.0, but weights are a research preview for non-commercial use only (repo license section)
- · Evaluations author-reported only; no independent benchmark numbers to cite