At a Glance
| Vendor | OpenAI |
| Family | GPT-6 series |
| Launched | OpenAI announced GPT-6 Astra on September 3, 2026, releasing it first to a limited set of organizations and making it broadly available from September 4. The launch followed a delay after unsanctioned cyber incidents involving OpenAI agents in July 2026, which OpenAI used to add safeguards. It is the first GPT-6 model and succeeds GPT-5.6 Sol as OpenAI's flagship. |
| Context window | 1,050,000 tokens with 128,000 maximum output tokens. Knowledge cutoff is April 30, 2026. Input is text and image; output is text only. Reasoning effort runs from low through medium, high, and xhigh to max. |
| Pricing | $10 per million input tokens, $1 per million cached input tokens, and $50 per million output tokens. Prompts above 272,000 tokens are billed at twice the input and cache rates and 1.5 times the output rate. Batch runs at half price and a Fast mode is available at a premium. This is five times the input price of GPT-5.6 Sol. |
| Access channels | OpenAI API (gpt-6-astra), ChatGPT Plus, Pro, Business, and Enterprise with staged rollout, Codex, and Amazon Bedrock. OpenAI's system card classifies it as the first broadly deployed OpenAI model to reach the Critical level of cybersecurity capability under its Preparedness Framework, and some advanced cyber features were held back at launch. |
Notable Benchmarks
Reported results include 57.7 percent on Terminal-Bench 4.0 (Fable 5.1: 55.8), 72.6 percent on OSWorld 2.0, 96.0 percent on GPQA Diamond, and 57.2 percent on Humanity's Last Exam with tools, below Fable 5.1's 65.0. On OpenAI's MRCR v2 long-context test it reports 100 percent at 256K to 512K tokens and 96.3 percent at 512K to 1M, against 73.8 percent for GPT-5.6 Sol in the longest band. A widely repeated ARC-AGI-3 score near 99.9 percent depends on a stateful, expensive test harness; reported scores through plain API calls are far lower, so treat it as a harness result rather than a general capability claim. Most scores were run at maximum effort.
Strengths
Long-context retention, computer and browser use, long-horizon agentic work, and science. In Codex, Astra keeps searchable notes across context windows instead of repeatedly compressing history into one summary.
Limitations
Expensive relative to GPT-5.6 Sol, with a steep surcharge beyond 272K tokens. OpenAI's own system card says Astra's reasoning is less monitorable than Sol's, a point safety researchers have criticized. Free and Go ChatGPT users do not get it at launch.
Brand-Visibility Implications
For brand visibility the practical question is which ChatGPT users see Astra. It is limited to paid plans at launch, so most ChatGPT answers about brands still come from the GPT-5.6 family, while the paid business users most likely to research vendors in depth now get a model with a much stronger long-context and browsing profile. Expect more thorough multi-source research answers on B2B and high-consideration queries, where a brand's full documentation and third-party coverage get read rather than skimmed. Segment ChatGPT tracking by model and plan where possible. See ChatGPT citation patterns and ChatGPT brand visibility.
How Presenc AI Tracks This Model
Presenc AI monitors brand visibility on OpenAI's GPT-6 series as part of continuous multi-platform AI visibility tracking. We sample OpenAI GPT-6 Astra across representative prompt sets daily, compare against competitor performance on the same prompts, and flag material mention-rate changes so brand teams can respond quickly when AI representation shifts.