GPT-6 Astra Frontier Model
OpenAI unveiled GPT-6 Astra on 3 September 2026, describing it as its most capable and aligned model so far, with gains across professional work, factuality, coding, computer use and safety alignment. It shipped first to a limited set of organisations, then to ChatGPT Plus, Pro, Business and Enterprise plus the OpenAI API and AWS
Tool Interface
Interactive tool will be available soon
Features
- ✓ Positioned by OpenAI as the strongest software-engineering model yet, with long-horizon coding lifted alongside it
- ✓ Computer use runs nearly 2x faster, and harness optimisations also speed up equivalent tasks on existing models by around 60%
- ✓ Staged rollout: a limited set of organisations first, then ChatGPT Plus, Pro, Business and Enterprise
- ✓ The same capability reaches the OpenAI API and AWS; enterprise plans stay off by default until an admin enables it
- ✓ Cyber-sensitive capabilities sit behind a trusted-access programme, so most users get the hardened public build
How to Use
- Read the official launch page at https://openai.com/index/gpt-6-astra/ for capabilities and rollout details
- Pick GPT-6 Astra in ChatGPT if your plan already has it, and trial it on one real work task first
- Developers should switch the model id to gpt-6-astra and keep the model name configurable for easy canary rollback
- Admins enable Astra access in the console before opening it up team by team
FAQ
What is GPT-6 Astra?
It is OpenAI's new frontier model, unveiled on 3 September 2026. OpenAI describes it as its most capable and aligned model so far, with improvements spanning computer use, coding, scientific reasoning, cybersecurity, professional work and long-context retrieval. The launch began as a limited preview; the official page is https://openai.com/index/gpt-6-astra/ and remains the source of truth for capabilities, benchmark data and safety detail.
When can I use it?
Per OpenAI, it went to a limited set of organisations on 3 September and then to all ChatGPT Plus, Pro, Business and Enterprise users over the following days, as well as through the OpenAI API and AWS. Enterprise access is off by default until an admin turns it on.
What changed most versus the previous release?
Two things stand out. First, software engineering: OpenAI calls it its best model yet for that work. Second, computer use is nearly twice as fast, and the harness optimisations also apply to existing models, speeding up equivalent tasks on older models by roughly 60%.
Why are some capabilities unavailable?
Cyber-sensitive capabilities were placed behind a trusted-access programme for vetted organisations, and the public build rejects or reroutes certain prompts. That matches OpenAI's broader move to slow releases and strengthen frontier safeguards earlier this year.
What must developers change?
The model id is gpt-6-astra, called through the OpenAI API or AWS. Keep the model name in configuration rather than hard-coded, start with shadow traffic and a narrow canary, and only widen once your evals hold so you can roll back instantly.
What about pricing and availability?
According to the official launch page, OpenAI API standard pricing is $10 per million input tokens and $50 per million output tokens, with separate cache read and write rates. The API also offers a Fast mode at up to twice the speed of standard processing, priced at twice the standard rate. Availability and plan thresholds move with each rollout stage, so check OpenAI's official pricing and release notes for the latest.