Every so often a new model shows up, aces every test in the room, and casually asks if there's an extra-credit assignment. Meet GPT-6 Astra, OpenAI's newest release, which posted scores so clean you'd assume grading was optional.

The Report Card Nobody Believes

OpenAI rolled out GPT-6 Astra on September 3, calling it the most intelligent and aligned model it has ever shipped. It's landing first with a limited set of organizations before spreading to ChatGPT Plus, Pro, Business, and Enterprise users, plus the API and AWS, over the following days.

The benchmark numbers are the kind that make you squint: 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and a perfect 100% on ExploitBench. API pricing runs $10 per million input tokens and $50 per million output tokens, with a 1.05-million-token context window attached.

The Part That Should Actually Get Your Attention

Buried under the highlight reel is a heavier detail: OpenAI says Astra is the first model it has broadly deployed that hits the "critical" tier of cybersecurity capability under its own preparedness framework. That's not marketing copy, that's the company flagging its own creation as genuinely dangerous in the wrong hands.

Access to Astra's sharpest edges is being gated first to companies in OpenAI's application-based cybersecurity program, which tells you this isn't just a smarter chatbot rollout — it's a model OpenAI itself thinks needs guardrails before the general public gets the keys.

A model this capable is exciting right up until you remember capability cuts both ways — for the defenders and the people trying to get past them.

If you're a business trying to figure out where a model like Astra actually fits into your workflows without opening a new attack surface, that's exactly the kind of AI integration conversation James and the WTK team have every week — let's talk it through.

Source: CNBC