What happened
OpenAI released GPT-6 Astra on Wednesday, September 3, positioning it as the clearest step yet toward AGI. Per CryptoBriefing, the model solved ten open problems in mathematics during a public $2,000 bounty run, a benchmark the company had used privately for months. Astra also crossed the Critical threshold on OpenAI's internal cybersecurity capability scale, the highest tier the lab defines under its Preparedness Framework. It is the first frontier model from the company to trip that classification, which by OpenAI's own policy requires additional deployment safeguards before broad access.
The launch was announced by OpenAI on Wednesday afternoon US time. Pricing, tool-use limits, and enterprise availability were not fully detailed in the initial announcement. CryptoBriefing first reported the math-bounty result and the Critical cybersecurity flag.
Why it matters
Two things stand out. The math benchmark isn't a synthetic test. Open problems have resisted proof for years, and a model closing ten of them in a single bounty round is a jump, not an incremental gain. The Critical cybersecurity rating is the bigger story for policy. Under OpenAI's Preparedness Framework, Critical is the tier at which the company has publicly said it would not ship without mitigations. Astra shipping anyway signals either that those mitigations are in place or that the framework is being interpreted more permissively than markets assumed.
The headline reads bullish. The governance picture is more complicated.
Market impact
There are no live prints in the provided data block, so we won't call bar-by-bar moves. The read-across is straightforward. AI-adjacent tokens -