“OpenAI has released GPT-6 Astra, its most powerful AI yet, capable of using computers and browsers, handling long multi-step tasks, creating documents and presentations, and building software autonomously" Secondary claims in the post (also investigated): (a) "Astra scored 97.6% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and 100% on…”

The release is real and confirmed on OpenAI's own channels. OpenAI's news index lists "GPT-6 Astra: A new generation of intelligence" dated Sep 3, 2026, alongside a "Safety overview: GPT-6 Astra" and a "GPT-6 Astra System Card" on the same date. The API documentation lists the model as built for "complex reasoning, coding, computer use, research, and document creation," with reasoning.effort supporting low, medium, high, xhigh, and max. The advertised capabilities in the claim track OpenAI's own wording closely. OpenAI states that Astra "combines the intelligence required for complex problems with the ability to carry out multistep workflows and produce polished documents, spreadsheets, and presentations," and that it is "our best model for adhering to existing templates and producing slides." The announcement says Astra "is state-of-the-art on computer use, browsing, software engineering, cybersecurity, science, and professional work," that it "saturates FrontierMath Tier 4 with a 98% score," and that it "also saturates ARC-AGI-3 with a 99.9% score and ExploitBench with a 100% score." The three headline numbers in the caption are OpenAI's own published figures, but the ARC-AGI-3 number is harness-dependent, and the benchmark's own operator published a very different result the same day. ARC Prize reports that "GPT-6 Astra scores 62.7% for $26K on ARC-AGI-3 Semi-Private with our Standard harness... and 99.9% for $19K with a Provider Adapter harness," where the Provider Adapter "preserves opaque reasoning state between requests and uses compaction for longer conversations, allowing the model to reuse prior work." ARC Prize's own summary describes Astra as scoring "63% on ARC-AGI-3, 99% via a new provider adapter harness." The widely shared comparison was also not like-for-like: "The figure that spread was 99.9% against GPT-5.6 Sol's 7.8%. Those are not the same test. Astra's 99.9% came from the Provider Adapter; Sol's 7.8% came from the standard harness." Independent aggregate evaluators disagree with the "most powerful" framing on general intelligence. The Decoder reports that Epoch AI places Astra clearly first with 169 points while "Artificial Analysis rates it no better than its predecessor," at 61 points, level with GPT-5.6 Sol and behind Claude Fable 5.1 at 66. Artificial Analysis' own writeup reports "a 6 point gain in Humanity's Last Exam" offset by "a drop of ~80 Elo points in GDPval-AA v2... measuring economically valuable tasks across 44 occupations," plus "2-3 point regressions on other evaluations." On its Coding Agent Index, Artificial Analysis found "GPT-6 Astra equals Fable 5 at less than half the cost," with "Fable 5.1 in Claude Code" leading at 70. A later index revision moved Astra up: Artificial Analysis raised its Intelligence Index to version 4.3, where GPT-6 Astra (max) and Claude Fable 5.1 both score 53 and share the top of the ranking. On availability at the time the post was published, the post overstated day-one access. OpenAI's announcement says Astra "is rolling out today to a limited set of organizations and over the coming days will become available to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock." Trade press reported that only organizations enrolled in OpenAI's Daybreak cybersecurity program could initially access the model, while "Plus, Pro, Business, and Enterprise ChatGPT subscribers, along with developers using the OpenAI API, were left out." Altman acknowledged: "First, sorry for the messy rollout." Access widened the following day: Altman posted that "GPT-6 Astra is now available to all Pro, Enterprise, and Business Premium users in Work/Codex, and is available in the API. We will start rollout to Plus and Business users next." On the AGI framing, the quote is real but the attribution of the judgment to OpenAI as an institution is not supported. Axios reports that Brockman called Astra a "generational leap" and said it could eventually be seen as the arrival of AGI, and that he "personally believes OpenAI has reached AGI, while leaving users to decide." He closed the press briefing with "Welcome to the AGI era." The benchmark operator whose score anchored the AGI narrative declined to endorse it: ARC Prize's own position is that "the benchmark's authors say they are not claiming AGI." Chollet does not call the result proof of AGI, though he describes progress running "twice as fast" than expected and is moving up his forecast.