OpenAI’s Sora Could Be Here This Week

Big news! I’m hearing that an updated version of OpenAI’s video generation model, Sora, is coming this week. This is likely the culmination of the work the company has been doing to train a new version of the model that can quickly generate clips that are higher in quality and longer than the ones it demonstrated earlier this year, as we reported in September. 

That’s bad news for video AI startups like Runway and Pika that develop competing products as well as Chinese video apps that have taken advantage of OpenAI’s slowness to release Sora. Of course, who knows how long it’ll take for OpenAI’s video generation model to become widely available after it’s announced.

The upcoming Sora release is part of a deluge of new products coming from OpenAI. Last week, the company kicked off 12 days of shipping products, or “shipmas,” by releasing a “full” version of its o1 “reasoning” model and a new version of that model for ChatGPT that is supposed to answer difficult questions by spending more time calculating the answer. The new high-end version for ChatGPT isn’t cheap: using it will cost $200 per month, though that will hardly come as a surprise to readers of this newsletter.

The average American adult already spends $91 per month on subscriptions, so spending another $200 on a ChatGPT tier won’t appeal to everyone. The target demographic will be folks such as the researchers at the Lawrence Livermore National Laboratory who are using o1-preview and o1-mini to help them with nuclear fusion projects.