OpenAI released GPT-6 Astra, the newest version of its GPT series, as a limited preview on September 3.
The model is rolling out to a limited set of organisations first, before becoming available to all ChatGPT Plus, Pro, Business and Enterprise users, as well as through the OpenAI API, Microsoft Azure and AWS Bedrock.
Astra is state-of-the-art on computer use, browsing, software engineering, cybersecurity, science and professional work.
The model achieves a score of 98% on FrontierMath Tier 4, a demanding mathematics benchmark, and 99.9% on ARC-AGI-3, a test of general reasoning.
It also scores 100% on ExploitBench, a benchmark measuring cybersecurity exploit capability.
So, some of the commentary, along with from OpenAI president Greg Brockman's launch keynote, suggest we are the precipice of AGI, artificial general intelligence.
Safety gating
GPT-6 Astra is the first OpenAI model to cross the "Critical" cybersecurity threshold in the company's Preparedness Framework, and the rollout is has been staged for that reason.
In testing, the model developed working exploits for hardened browsers and operating systems, and it surfaced two previously unknown vulnerabilities.
The publicly available version refuses advanced offensive security work such as exploit discovery, with less-restricted capabilities routed to vetted defenders through a programme called Daybreak Blue.
Anthropic takes a broadly similar approach with its own most capable systems.
Anthropic's current lineup includes Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5, alongside a newer tier called Mythos, which sits above Opus.
The first Mythos-class model, Claude Mythos Preview, is not yet available to the public and is being used by a small number of trusted organisations under a scheme called Project Glasswing.
Two further Mythos-tier models, Claude Mythos 5.1 and Claude Fable 5.1, share the same underlying model, though Fable carries additional safety restrictions covering biology, cybersecurity and AI research capabilities.
That mirrors OpenAI's approach of gating the riskiest capabilities of its most powerful model behind vetted access rather than releasing them broadly.
Price and specification
GPT-6 Astra's API pricing runs at two and a half times the previous flagship model on both input and output tokens.
Standard API pricing starts at ten dollars per million input tokens and fifty dollars per million output tokens.
Anthropic has not disclosed pricing for its Mythos-tier models, as they remain restricted to trusted partners rather than being sold commercially.
GPT-6 Astra offers a context window of just over one million tokens, matching the scale Anthropic has offered with recent Claude models, and a maximum output of 128,000 tokens.
For most users on standard paid plans, the practical differences between GPT-6 and Claude will come down to task performance, pricing and how quickly each company widens access to its top-tier models.
