OpenAI · Model release
GPT-6 Astra: OpenAI's computer-use flagship arrives gated
OpenAI's first GPT-6-family model is a computer-use flagship — and the first model it gates at the Preparedness Framework's 'Critical' cybersecurity threshold.
What changed
- Trained on >100,000 GPUs at Stargate — OpenAI's largest training run to date — and rolled out over several days from 2026-09-03 across ChatGPT, the API, Azure, and Bedrock.
- Priced $10/$50 per MTok (≤263K context) rising to $20/$100 above it, with cached input at $1/MTok and a 2x-price Fast mode.
- Benchmarks at launch: 61.2 on the AA Intelligence Index v4.1.1 (below Fable 5.1's 65.7), a self-reported 74.1% on DeepSWE 1.1, and 72.6% on OSWorld 2.0 offline.
Best fit
- Computer-use agents
- Agentic coding
- Hard reasoning with vision
What to be careful about
- Gated at the 'Critical' cyber threshold; Enterprise tenants get it disabled by default and eligibility checks apply.
- It is not the AA-index leader — Fable 5.1 still tops that board at launch.
- DeepSWE and OSWorld figures are OpenAI's own until independent boards pick the model up.
Dataset snapshot: GPT-6 Astra is also represented in the checked comparison dataset. Prices and benchmark figures carry their own field dates.