VPN World • Technology

OpenAI Unveils GPT-6 Astra, Its Most Powerful Model Yet, as a Trillion-Dollar IPO Looms

OpenAI released a new flagship on Thursday and called it the world’s most intelligent and aligned system. It is the company’s first model gated for its own cyber capabilities, and it arrives as OpenAI prepares to go public. Independent testers, though, still rank it behind Anthropic’s Claude on general intelligence.

By Lance Davis • Vincypowa News • Power to the People

OpenAI launched GPT-6 Astra on September 3, 2026, describing it as the most intelligent and aligned model in the world and the new state of the art for computer use, browsing, software engineering, cybersecurity, science and professional work. Company president Greg Brockman used the briefing to make a larger claim. He said the model could qualify as artificial general intelligence, and suggested it is not unreasonable to feel the industry has now entered what he called the AGI era, while conceding that AGI itself remains a fuzzy and ill defined threshold.

Astra is OpenAI’s first new general-purpose flagship since the GPT-5.6 family, Sol, Terra and Luna, shipped in July. The reveal comes as OpenAI moves toward one of the largest stock market debuts in history, and the timing is not a coincidence.

What Astra is

OpenAI leaned on a wall of benchmark results. The company says Astra reaches about 97.6 percent on FrontierMath Tier 4, a research-level mathematics test it rounds to 98 percent in its own headline, and says the model has already helped solve long-standing open problems in mathematics. On ARC-AGI-3, a test built to measure reasoning on unfamiliar problems, OpenAI reports 99.9 percent using its own enhanced tool harness and about 66 percent on the benchmark’s standard harness, against 7.8 percent for GPT-5.6 Sol and roughly 30 percent for Anthropic’s Claude Opus 5. On cybersecurity, Astra posted a perfect score on ExploitBench.

The headline feature is speed at computer use. Brockman said the model can move through spreadsheets, fill out forms and navigate web pages at what he called superhuman speed, and OpenAI framed Astra as a system you can hand multistep work to with more confidence in its judgment.

The benchmark picture, with a caveat

The gaudy numbers deserve context, and the most important context is this: OpenAI and Anthropic launched their newest models about 48 hours apart, and neither benchmarked against the other. Anthropic shipped Claude Fable 5.1 on September 1 and measured it against GPT-5.6 Sol, because Astra did not exist yet. OpenAI shipped Astra two days later and measured it mostly against its own predecessor. Every head-to-head figure in circulation is therefore stitched together from two companies talking past each other.

Independent trackers tell a cooler story than the launch post. On the Artificial Analysis Intelligence Index, a general-intelligence measure, Astra scored 61, level with GPT-5.6 Sol and five points behind Claude Fable 5.1 at 66. On coding, the models are close. Astra scored 67 on the Artificial Analysis Coding Agent Index, roughly equal to Claude Opus 5 and Claude Fable 5, while Fable 5.1 led that index at 70. Astra’s real edge on coding is cost, not raw score: the trackers found it can match Fable 5 on the same task at less than half the price, driven by large gains in how few tokens it burns.

Safety and the “Critical” rating

Astra is the first OpenAI model rated Critical for cybersecurity under the company’s Preparedness Framework, and that rating held up the launch. Advanced cyber capability is gated to vetted users through a program OpenAI calls Daybreak, while standard ChatGPT and API access is designed to refuse high-risk cyber work. OpenAI also said the model shows better alignment than earlier versions, including fewer attempts to push past authorized targets on tasks it cannot complete. The release lands amid rising scrutiny of whether frontier labs can keep control of what they are building.

Price and rollout

List pricing is $10 per million input tokens and $50 per million output tokens. That matches Anthropic’s Claude Fable 5.1, which carries the same $10 and $50 rates, though the two differ underneath: OpenAI lists Astra cache reads at $1 per million while Anthropic lists Fable 5.1 at 25 cents. Claude Opus 5, at $5 and $25, remains half Astra’s headline rate. Access is phased, starting with a limited set of organizations and expanding to ChatGPT Plus, Pro, Business and Enterprise users, the API and Amazon Bedrock in the coming days.

Why it matters now

Anthropic had been taking enterprise and coding mindshare with its Claude Fable and Mythos models, and it also carries a higher private valuation. OpenAI wants that narrative back before it faces public investors. The company filed confidentially for a U.S. IPO in June 2026. Its last private valuation was about $852 billion, set in a funding round earlier this year, and it has targeted a listing above $1 trillion, with Sam Altman reported to treat any lower figure as a non-starter. The timing is still fluid. Some reports point to a listing as early as late 2026, while others say 2027 after bankers cautioned that markets may not support the price OpenAI wants.

Astra is two things at once: a product launch, and a pitch to investors that OpenAI still sets the frontier.

Why it matters here

For St Vincent and the Grenadines, the frontier fight plays out one tier down. The cyber-grade version of Astra is locked to vetted users, and its API rates are priced in U.S. dollars at the top of the market, out of reach for most local budgets. What reaches Vincentian students, small businesses and newsrooms is the consumer ChatGPT tier, where faster computer use and stronger document and spreadsheet work could sharpen everyday tools. The steady fall in cost per task across these models is the trend worth watching locally, because it decides when this kind of capability becomes affordable in Kingstown rather than only in Silicon Valley.

VPN verifies technology claims against independent testing, not company marketing. Figures in this report were checked against OpenAI’s launch materials and third-party trackers as of publication. Power to the People.