Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• Molmo2's 8B model outperforms other open weight and data models on short videos, counting, and captioning tasks.result 90%
• Molmo2 significantly outperforms existing open-weight models like Qwen3-VL on video counting tasks (35.5 vs 29.6 accuracy).result 95%
• Molmo2 surpasses proprietary models like Gemini 3 Pro on video pointing tasks (38.4 vs 20.0 F1).result 95%
• Molmo2 introduces 7 new video datasets and 2 multi-image datasets collected without using closed VLMs.method 90%
• Bi-directional attention on vision tokens and a novel token-weight strategy improves model performance.technical 85%
• Molmo2 demonstrates exceptional new capabilities in point-driven grounding in single image, multi-image, and video tasks.technical 90%
• Handling long-duration videos with complex scenes might present challenges for Molmo2.limitation 80%
• Molmo2 has the potential to replace proprietary video-language models by offering similar or better performance while being fully open-source.market 75%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card