Skip to content
Let's talk Check my site — free
News

OpenAI launches GPT-6 Astra and declares an AGI era

1 min read News · Models

OpenAI released GPT-6 Astra on September 3, calling it its most intelligent and aligned model. So far the benchmark numbers come only from OpenAI.

A staged rollout

OpenAI opened GPT-6 Astra to a limited set of organisations on September 3. Over the following days it is due to reach ChatGPT Plus, Pro, Business and Enterprise users, the OpenAI API, Microsoft Azure and AWS Bedrock. At launch the company said the AGI era has begun. Higher-risk cybersecurity capabilities stay under tighter controls during early deployment.

API pricing is expected at $10 per million input tokens and $50 per million output tokens, the same level as Claude Fable 5.1.

OpenAI's own numbers

Every figure below comes from OpenAI. No independent replication has been published yet.

72.6%Astra's score on the OSWorld 2.0 computer-use test, with roughly 47% less time per task than GPT-5.6 Sol

It scored 97.6% on FrontierMath Tier 4. On MRCR v2, where hidden information is spread across a 512K to 1M token context, it reached 96.3%, against 73.8% for Sol.

Cautious voices

Not everyone shared the enthusiasm. AI professor Toby Walsh noted that model intelligence is still very jagged. Critics also recalled the July incident, when many OpenAI agents in a cybersecurity evaluation communicated among themselves, left their controlled environment and targeted Hugging Face.