Archive · 7 Sep 2026
OpenAI Launches GPT-6 Astra, Claims Edge Over Claude
OpenAI has released a new artificial-intelligence model called GPT-6 Astra.
It is designed to help with coding, computer tasks, science and cybersecurity.
OpenAI says Astra is faster and more accurate than its earlier model for many everyday computer jobs.
The company also says Astra performed better than Anthropic’s Claude Fable 5.1 on several tests.
However, Claude Fable 5.1 scored higher on one broader intelligence measurement published by OpenAI.
Astra is being introduced first to some organizations and then to paid ChatGPT users.
People can also use it through the OpenAI API and certain cloud services.
Because Astra is powerful in cybersecurity, its public version refuses some advanced hacking-related requests.
OpenAI says it may allow more cybersecurity capabilities for approved users later.
GPT-6 Astra is rolling out to selected organizations before broader access for paid ChatGPT users and through major cloud platforms.
OpenAI priced Astra’s API at $10 per million input tokens and $50 per million output tokens.
The company says Astra improves computer use, coding, cybersecurity, science and professional work, including faster task completion.
OpenAI said Astra crossed its internal “Critical” cybersecurity threshold and that the public version blocks advanced exploit-building requests.
OpenAI’s benchmarks show Astra ahead of Claude Fable 5.1 in several tests, although Fable 5.1 led on a broader intelligence index.
- Who
- OpenAI launched GPT-6 Astra, with comparisons against Anthropic’s Claude Fable 5.1.
- What
- A new flagship AI model was released with claimed improvements in coding, computer use, cybersecurity, science and reasoning.
- When
- The rollout began with selected organizations, with wider availability planned over the following days.
- Where
- Access is planned through ChatGPT, the OpenAI API, Microsoft Azure and AWS Bedrock.
- Why
- OpenAI said Astra was developed to provide stronger performance on professional, technical and reasoning tasks.
OpenAI’s Performance Claims
Comparative Caveats
Benchmark leadership
OpenAI’s Performance Claims
OpenAI’s published figures show Astra outperforming Claude Fable 5.1 on several coding, science, professional-work and 3D-modelling tests.
Comparative Caveats
The comparison is based on OpenAI’s own benchmark tables, and Claude Fable 5.1 scored higher on the broader intelligence index cited in those tables.
Reasoning performance
OpenAI’s Performance Claims
OpenAI said Astra achieved perfect or near-perfect results on several advanced reasoning benchmarks and surpassed earlier OpenAI models and Claude models on key tests.
Comparative Caveats
The article does not provide independent verification of those claims, so the reported advantage remains based on company-published results.
Cybersecurity access
OpenAI’s Performance Claims
OpenAI said Astra’s stronger cyber capabilities could eventually be made available to vetted users through its Daybreak programme.
Comparative Caveats
The publicly released model restricts advanced cybersecurity requests, including requests to build proof-of-concept exploits, because of the risks associated with its capabilities.
- Model
- GPT-6 Astra
- API pricing
- $10 per million input tokens and $50 per million output tokens
- Context window
- 1,050,000 tokens
- Maximum output
- 128,000 tokens
- Computer-task speed
- OpenAI said Astra completed tasks about 47% faster than its predecessor
- Cybersecurity status
- OpenAI said Astra crossed its internal “Critical” threshold
- Availability
- Rolling out to selected organizations, then ChatGPT Plus, Pro, Business and Enterprise users