OpenAI · GPT-6
GPT-6 Astra
OpenAI’s new flagship for research, coding, computer-use and long-running agentic work. It began rolling out on 3 September 2026, first to a limited set of organisations and then to ChatGPT Plus, Pro, Business and Enterprise, the OpenAI API and AWS.
Treat the capability claim as a reason to test, not to delegate. Its benchmark scores and most-intelligent framing are OpenAI’s own, its comparison to Claude was drawn against Fable 5 rather than the current Fable 5.1, and the launch that widens what the model will attempt on its own is the same one OpenAI rated Critical on cyber capability. Run one real task beside another flagship and keep a human checkpoint on anything you could not easily undo.
Use it for
- Research, coding and computer-use work that runs several steps on its own
- Document, spreadsheet and presentation work with a clear acceptance check
- Long-running agentic tasks where a person still owns the result
Do not reach for it where the missing piece is your own source, judgement or sign-off rather than more capability, and do not read “can do anything you can do on a computer” as permission to hand over work you cannot check.
Effort without the jargon
The API exposes adjustable reasoning effort at low, medium, high, xhigh and max; start at a sensible middle level and raise it only when the task, not an unclear prompt, is why a shallower pass failed.
Advanced recordCost, limits, privacy and operations
Astra and Claude Fable 5.1 now share the same headline token price, so price is not what separates them. What differs is cache economics and how much work each will attempt on its own; test on your real task and read the actual bill.
- API cost
- Officially disclosedOpenAI API price: US$10 per million input tokens and US$50 per million output tokens, with cached input at US$1 and cache writes at US$12.50 per million tokens. The headline input and output rates match Claude Fable 5.1; the cache economics differ, since Fable 5.1 reads cache at US$0.25 per million tokens against Astra’s US$1.
- Context window
- Officially disclosed1,050,000 tokens
- Maximum output
- Officially disclosed128,000 tokens
- Knowledge cutoff
- Officially disclosed30 April 2026
- Recorded inputs
- Text · Images
- Recorded output
- Text
Data handling is a product decision
Depends on surfaceRetention and training rules belong to the ChatGPT plan, API account and workspace, not the model name alone. Astra rolls out across ChatGPT, the OpenAI API and AWS, which are different surfaces; check the exact one before adding sensitive material.
Operational constraints
- The model is mid-rollout: it reached a limited set of organisations first, with wider ChatGPT, API and AWS availability following in the days after 3 September 2026, so access depends on plan and cohort.
- Benchmark scores and the most-intelligent framing are vendor self-reported and were not independently replicated at review time; treat them as claims.
- OpenAI rates Astra at the Critical cybersecurity capability level under its Preparedness Framework, the stated reason for the phased rollout, which warrants a security review before broad use.
- Cloud availability beyond AWS was not confirmed on a directly readable OpenAI page at review time.
Unknown means the cited official record does not disclose a safe value. It is not an estimate. Prices are provider list prices in USD where stated and can change before this record's review date.
Keep the evidence separate
OpenAI describes GPT-6 Astra as its most intelligent and aligned model and says anything you can do on a computer, Astra can do for you, aimed at research, coding, computer use and long-running agentic work. It also says Astra is the first model it has rated at the Critical cybersecurity capability level under its Preparedness Framework, which is the stated reason the rollout is phased.
No sufficiently useful independent note has been added yet. That absence is not filled with our guess.
No Human Bit result is claimed; use this as evaluation guidance only.
Not scheduled
No Human Bit result is claimed for this model yet.