- Published on
First Demo Samples of OpenAI Astra Leak, Generating GTA 2-like Game in One Interaction
- Authors
- Name
- DeepHorus
OpenAI is accelerating the release process for its next-generation AI model, Astra. Tech media outlet testingcatalog disclosed in a blog post on August 29th that OpenAI has added a new testing phase named "mozaik-alpha-fdm" for Astra, aiming to expand the model's internal testing scope. Astra is a model series developed internally by OpenAI, with its codename continuing the company's tradition of celestial naming, following previous ones like Sol, Terra, and Luna. This codename, meaning "star" in Latin, also hints at its potential significant implications.
According to reports from multiple media outlets, including The Information, the Astra model has demonstrated major breakthroughs in multiple frontier scientific fields such as mathematics, physics, biology, medicine, chemistry, and cybersecurity. Rumors suggest that the model has even solved several long-standing, unsolved high-order mathematical problems. Currently, Astra has not been officially confirmed as "GPT-6" by OpenAI, but the industry and media widely speculate that it is likely the highly anticipated GPT-6, or one of its candidate names. Analysis suggests its parameter scale could reach an astonishing 10 trillion level.
The leaked first batch of demonstration samples intuitively showcases Astra's powerful capabilities in content generation. Test content shared by netizens indicates that the model has particularly enhanced the efficiency of code writing and visualization software creation. Most notably, through a single conversational interaction, Astra successfully generated a game program similar to the classic game "Grand Theft Auto 2," while also creating detailed websites, 3D objects, and voxel environments. This demonstrates significant progress in the model's ability to understand complex instructions and convert them into executable code.
However, current tests also reveal some characteristics of Astra at this stage. Testers reported that Astra's demonstration was in a "zero-shot testing" state and used Max reasoning intensity. This means that the time required for a single inference is much longer than that of the current GPT-5.6 Sol model. This could be a trade-off made to exchange for higher quality and the ability to complete more complex tasks, and it also foreshadows that the final product may seek a new balance between performance and efficiency.
If the efficient development capability demonstrated by Astra can be carried over to the final released product, its impact will be profound. Developers, designers, and entrepreneurs who need to quickly prototype could become the primary beneficiaries. They might be able to use Astra to transform conceptual ideas into interactive prototypes in an extremely short time, thereby greatly accelerating the product innovation and iteration cycle.
Currently, the specific release date and final parameter configuration of the Astra model remain unclear. OpenAI is polishing the model's performance by continuously expanding internal testing. The leakage of this demonstration sample has undoubtedly raised industry expectations for the next-generation general artificial intelligence model. The market will closely monitor every step of Astra's progress from the testing phase to the public release process.
