What happened
On September 3, 2026, OpenAI introduced GPT-6 Astra, which it calls "the world's most intelligent and aligned model". It first rolled out to a limited set of organizations and, over the following days, to ChatGPT Plus, Pro, Business and Enterprise, the OpenAI API (as gpt-6-astra), Microsoft Azure and Amazon Bedrock. AWS announced general availability on Bedrock in September.
OpenAI's reported numbers
According to OpenAI (not yet independently verified):
- 98% on FrontierMath Tier 4;
- 99.9% on ARC-AGI-3;
- 100% on ExploitBench, an offensive-security benchmark;
- state of the art in computer and browser use.
Pro, Business and Enterprise plans also get GPT-6 Astra Pro. In Enterprise workspaces access is off by default and must be enabled by an admin.
Follow-ups the same month
On September 22 the family added GPT-6 Sol and GPT-6 Luna, and on September 29, at DevDay, GPT-6.1 Sol arrived, which OpenAI says delivers near-Astra intelligence at a fifth of the token price.
Why it matters
A model that maxes out an exploitation benchmark is a powerful tool for defense and potentially for attack. For businesses, that means reviewing who may use it, with which data and with what audit logging.
What to do now
- Enable Astra only after defining a usage policy and data classification.
- Compare cost: for many tasks GPT-6.1 Sol may be enough.
- Treat benchmarks as vendor claims and test on your own use cases.