Polymarket odds on OpenAI releasing its next major flagship model by August 15 soared from 10% to 42% this Thursday, before settling near 32%. The move was prompted by a post from X user @synthwavedd claiming that the AI lab is gearing up for an imminent release. The account - run by anonymous “tech, AI and politics nerd” Leo - has a strong track record of predicting frontier model release windows.
Leo claims that the current internal build of Astra, reportedly dubbed “mewfour” by staff, is a release candidate and could go live as early as next week.
OpenAI itself has made no announcement on when the new model, developed under the working title Astra, will be made available to the public. A report from The Information suggested that the company has not yet decided whether to release Astra as GPT-6 or as another iteration of the existing GPT-5 lineup.
News of its development first appeared in late July, when Sam Altman traveled to Washington to showcase its capabilities to lawmakers. Shortly after, OpenAI published a research post, covered by The Latent, that claimed the model had solved or made significant progress on ten open math problems during internal testing.
Anticipation around this previously unacknowledged new model quickly grew. Some read this as a marketing gambit by OpenAI.
The federal review that could push the date
The timeline offered by the X leak cites no sources and was made by an anonymous commentator. It should therefore be taken with a grain of salt, especially when recent developments in Washington may introduce delays and complications for some public model releases.
The government’s new federal review framework, developed following an executive order from President Trump signed on June 2, was presented to leading AI labs this Tuesday. Its guidelines give these firms the option to submit their new models for review by federal agencies up to 30 days before wider release, giving them time to assess their cyber capabilities.
Polymarket traders therefore have greater confidence in an August 31 release deadline, pricing the probability above 50%, as this would align with the full 30-day window if Astra were submitted around the August 1 announcement. However, this too is based on speculative assumptions.
It’s important to note that these markets have relatively limited volume. The August 15 contract shows just $7,000 traded, meaning that it reflects the opinions of a limited number of traders. It is therefore not a strong source of crowdsourced intel.
What a new GPT pretrain would mean
OpenAI will be hoping that Astra can close the narrow lead that rival lab Anthropic gained with the release of Claude Fable 5; Anthropic’s flagship model currently holds a one-point lead over GPT-5.6 Sol on the Artificial Analysis Intelligence Index.
In the alleged leak on X, Leo highlights that Astra is set to be "the largest model OpenAI have trained since GPT-4.5." This pretraining is the most expensive part of the development process, as the model must process a vast amount of material to build out its knowledge base.
If accurate, this would make Astra OpenAI’s largest new pretraining run since February 2025. At that time, GPT-4.5 proved so costly to operate that OpenAI removed the model from its developer API within five months.