Home / Technology / Ai2's Molmo 2: Video AI Gets Leaner
Ai2's Molmo 2: Video AI Gets Leaner
17 Dec
Summary
- Molmo 2 offers open-source video AI, challenging large proprietary models.
- It excels in video grounding and tracking, outperforming competitors.
- The model targets enterprises needing efficient video understanding.
The Allen Institute for AI (Ai2) has introduced Molmo 2, its latest open-source video model. This release aims to demonstrate that smaller, accessible models can effectively handle complex video understanding and analysis tasks, a domain previously dominated by large, proprietary systems.
Molmo 2 builds on its predecessor's strengths, expanding capabilities to video and multi-image comprehension. It supports various input formats, including video clips of different lengths, enabling tasks such as video grounding, tracking, and question answering. Ai2 highlights that a core design goal was to address the gap in open models regarding "grounding."
Despite its smaller size, Molmo 2 has shown competitive performance, even outperforming larger models like Google's Gemini 3 Pro on specific video tracking benchmarks. Ai2 positions Molmo 2 as an efficient alternative for enterprises focused on video analysis rather than generation, emphasizing its prowess in video grounding and counting.
