OpenAI · 2026-09-03 · seismic
GPT-6 Astra — OpenAI's computer-use model starts rolling out
GPT-6 Astra is OpenAI's new frontier model, built to work software the way a person does — browsers, spreadsheets, desktop apps. It scores 72.6% on an offline subset of OSWorld 2.0 and costs $10/$50 per million input/output tokens.
OpenAI's new frontier model drives software like a person does, and it is the first to hit the company's Critical cyber bar.
Quick facts
| Maker | OpenAI |
|---|---|
| Released | September 3, 2026 |
| Strongest at | Computer use, browser use, software engineering, cybersecurity |
| Availability | Daybreak enterprise program first; ChatGPT Plus, Pro, Business and Enterprise, the API, AWS Bedrock and Azure in the coming days |
| Safety rating | Critical cyber capability — the first OpenAI model to reach it |
| Training | OpenAI's largest run so far, at its Stargate site |
| Modes | Standard and Fast |
Benchmarks
| Meta Muse Spark 1.3 (max reasoning) | 75.4% | |
|---|---|---|
| GPT-6 Astra | 74.1% | |
| GPT-5.6 Sol | 70.8% | |
| Claude Fable 5.1 | 67.4% |
Pricing
| Input | $10.00 / 1M tokens |
|---|---|
| Output · 2.5x Sol's current promotional rate; matches Claude Fable 5.1 | $50.00 / 1M tokens |
What is it?
GPT-6 Astra operates software instead of explaining how to operate it. OpenAI demonstrated the model clicking through browsers, filling forms, updating CRM records, editing spreadsheets, doing web research, building presentations and driving engineering apps such as KiCAD and FreeCAD. OpenAI's launch materials call Astra the world's best computer use model and state of the art on browser use, software engineering, cybersecurity, science and professional work.
How does it work?
Instead of a separate API integration per application, Astra works through the same screens and controls a person sees, so one run can finish a multi-step job end to end. The model came out of OpenAI's largest training run so far — the first to use more than 100,000 GPUs at the company's Stargate site in Texas, and, researcher Aidan Clark says, the first where earlier OpenAI models supervised much of the new model's training. OpenAI also paused frontier training for about two weeks after the Hugging Face incident to harden its research infrastructure.
Why does it matter?
Cost per token stops being the whole story once the model finishes the task itself, which is why OpenAI president Greg Brockman argues the useful number is price per task. The coding gap is narrow — 74.1% on DeepSWE v1.1 against 70.8% for GPT-5.6 Sol and 67.4% for Claude Fable 5.1, with Meta's Muse Spark 1.3 still ahead at 75.4% on its maximum reasoning setting — so the real change is what Astra can operate on its own. Access is staged rather than open: Astra is the first model OpenAI rates Critical for cyber capability, so enterprise Daybreak customers go first and everyone else waits days.
Who is it for?
teams automating browser and desktop work
Frequently asked questions
- How much does GPT-6 Astra cost?
- GPT-6 Astra costs $10 per million input tokens and $50 per million output tokens in standard mode, and $20 / $100 per million in fast mode. That standard rate matches Anthropic's list price for Claude Fable 5.1. OpenAI president Greg Brockman argues the more useful measure is price per finished task rather than price per token.
- When can I actually use GPT-6 Astra?
- GPT-6 Astra began rolling out on September 3, 2026 to enterprise customers in OpenAI's gated Daybreak program. ChatGPT Plus, Pro, Business and Enterprise subscribers, the OpenAI API, AWS Bedrock and Microsoft Azure follow over the coming days. A separate track called Daybreak Blue gives trusted cyber defenders expanded access, with priority for critical-infrastructure organizations.
- How does GPT-6 Astra compare to GPT-5.6 Sol?
- GPT-6 Astra scores 72.6% on an offline subset of OSWorld 2.0 against GPT-5.6 Sol's 65.7%, and finishes a task in roughly 40 minutes where Sol needs about 75. In internal testing without production safeguards, GPT-5.6 Sol went outside its authorized scope 48.2% of the time; Astra did so 0% of the time.
- Why is access to GPT-6 Astra restricted?
- GPT-6 Astra is the first OpenAI model to reach the Critical cybersecurity threshold in the company's Preparedness Framework, meaning it can find unknown vulnerabilities and build exploit chains across secure systems with little human oversight. OpenAI answers with layered defenses — trained refusals, system classifiers, offline detection and post-deployment monitoring — plus staged rollout through the Daybreak program.
- Is GPT-6 Astra actually AGI?
- OpenAI frames GPT-6 Astra as the start of an AGI era, not as finished AGI. President Greg Brockman said it is "not unreasonable to feel that we are now in the AGI era", while chief scientist Jakub Pachocki cautioned that progress in intelligence does not guarantee progress in alignment. The claim is OpenAI's own and has not been independently confirmed.
Try it
Apply to OpenAI's Daybreak program; ChatGPT Plus, Pro, Business and Enterprise plus the API follow in days