Chinese Models Lead the Pack on Video AI, for Now

Before we get to today’s column, OpenAI is set to release its latest flagship model, GPT-5, later Thursday morning.

We may have taken some of the wind out of the launch’s sails thanks to our reporting in recent weeks on how GPT-5 came together and what you can expect from the model (sorry OpenAI!), but there’s still lots to look forward to with the release later today. Specifically, I’m interested in how OpenAI will show off the model’s improved coding capabilities, since that’s been an area that its rival Anthropic has so far excelled in.

Onto today’s column… 

Guess which video AI startups lead that increasingly hot sector? You might be surprised to learn it’s not U.S. firms such as Runway and Luma AI, both of which we reported this week are in talks to raise multi-billion-dollar funding rounds. Instead, the startups’ video models currently trail offerings from Chinese competitors. 

Artificial Analysis keeps a leaderboard of the best video models based on head-to-head comparisons of their outputs. Models developed by AI companies in China hold 14 out of the top 20 spots in the text-to-video category: first and third place go to video models from ByteDance, and social media company Kuaishou’s KlingAI models rank fifth and seventh. 

Major U.S.-made models are in the mix, too: different versions of Google’s video AI models are second, fourth and sixth, while OpenAI’s Sora is the 10th. (Google this week announced Genie 3, an AI model that can generate interactive videos.) But Runway’s Gen 3 Alpha model is at 22nd place and Luma’s Ray 1 takes 24th place—behind many Chinese models.