Eseandre Mordi
⬤ Fresh benchmark data from Artificial Analysis reveals LTX-2 now leads all open-weights video generation models. The model beat out Alibaba's Wan 2.2 A14B in head-to-head comparisons across both text-to-video and image-to-video tasks in the Artificial Analysis Video Arena.
⬤ The rankings show LTX-2 Pro (No Audio) and LTX-2 Fast (No Audio)—both from Lightricks—holding first and second place with ELO scores of 1,198 and 1,195. Alibaba's Wan 2.2 A14B sits in third with a score of 1,130. Other open models, including more Wan versions and Tencent's Hunyuan Video, trail behind, showing just how competitive the open video generation space has become.
⬤ LTX-2 first dropped in November and has since gone open source. The 19B-parameter base model and a smaller distilled version are both publicly available. These open-source versions are different from the commercial Pro and Fast API endpoints, which include extra pipeline improvements. You can grab the model weights on Hugging Face under the LTX-2 Community License, which allows commercial use up to a certain revenue cap but stops you from directly competing with Lightricks' own products.
⬤ These results highlight how quickly open-weights video models are advancing. LTX-2's edge over Alibaba's Wan series shows real improvements in quality and accessibility for text-to-video and image-to-video generation. As open models keep narrowing the gap with closed-source systems, benchmarks like the Artificial Analysis Video Arena are becoming the go-to standard for evaluating and choosing video generation tech.
Eseandre Mordi
