OpenAI has released its most powerful AI model yet, and this time the company isn’t being shy about what it thinks that means. GPT-6 Astra launched on Thursday, September 3, 2026, and OpenAI president Greg Brockman closed the press briefing with a simple line: “Welcome to the AGI era.”
That’s a bold thing to say out loud. AGI, or artificial general intelligence, is the industry’s holy grail — a system that can match or beat humans at most economically valuable work. OpenAI isn’t officially declaring that Astra is AGI. But Brockman told reporters it’s “not unreasonable to feel” the industry has entered that era, and that people may look back at this specific model as the turning point.
The Launch
GPT-6 Astra is OpenAI’s first sixth-generation model, replacing GPT-5.6 Sol at the top of its lineup. It’s rolling out in phases: first to a limited set of enterprise organizations in OpenAI’s Trusted Access Program, then to ChatGPT Plus, Pro, Business, and Enterprise users over the following days, and finally through the API and Amazon Bedrock.
OpenAI is pitching Astra as a jump across four areas at once: software engineering, scientific research, professional office work, and full computer use — meaning the model can click, type, and navigate software the way a person would, without constant hand-holding.
Key Details
Some of the numbers OpenAI and independent reporters are pointing to:
- ARC-AGI-3: Astra reportedly matched human action-efficiency on 96% of levels, which OpenAI is calling near human parity on that benchmark.
- ExploitBench: Astra scored 100%, compared to roughly 5.5% for GPT-5.6 Sol, its predecessor.
- Coding: Reports describe Astra completing tasks up to 47% faster than Sol while using fewer tokens.
- OSWorld (a computer-use benchmark): Astra reportedly scored 72.6%.
Astra is also the first OpenAI model to hit the company’s “Critical” cybersecurity risk threshold under its Preparedness Framework — meaning it can independently find unknown security flaws and build working exploits without step-by-step human guidance. During internal testing, the model reportedly found and chained together two previously unknown vulnerabilities on its own. OpenAI says it’s now disclosing those flaws to the affected software makers.
Because of that, two versions of Astra exist. The public version — the one landing in ChatGPT and the API — has its offensive cyber capabilities locked down with extra monitoring and guardrails. The full, unrestricted version stays inside Daybreak Blue, a separate program limited to vetted defense and security organizations.
Big Picture
If Astra can really operate a computer, write production-grade code, and hunt for security flaws at nearly an expert level, it could meaningfully change how we use AI at work—not just as a chatbot, but as something closer to a digital employee. That’s also exactly why OpenAI is framing this as an “AGI era” moment rather than just another model update.
But the framing is already drawing pushback. Some coverage points out that OpenAI hasn’t shown that Astra actually outperforms humans at most economically valuable work — which is the company’s own long-standing definition of AGI. A lot of Astra’s most impressive results also depend on the tools and permissions built around it, not the model alone.
Pricing and Availability
Astra is priced at $10 per million input tokens and $50 per million output tokens on the standard tier, with a faster processing option available at roughly double that rate. For comparison, GPT-5.6 Sol currently runs at about $4 and $20 under promotional pricing, so Astra costs roughly 2.5 times more per token.
No region-based restrictions apply to the standard version. It’s expected to reach users in the US and India, among other markets, as the ChatGPT rollout continues over the coming days. The API model ID being used is gpt-6-astra.
Official Statements
Brockman didn’t stop at “AGI era.” When asked directly if OpenAI was formally declaring AGI achieved, he reportedly said the term no longer represents a contractual milestone, referencing OpenAI’s earlier agreement with Microsoft. Instead, he described AGI as more of a “mission concept” than a hard technical benchmark. OpenAI’s own launch material describes Astra as its “most intelligent and aligned model” to date.
Industry Impact
Astra’s launch lands in the middle of a genuinely stacked week for AI releases, with Anthropic, Meta, and Google all pushing out updates to their own leading models around the same time. That pace is part of the story here: frontier labs are shipping faster than most businesses can evaluate, let alone adopt, what they’re releasing.
Final Thoughts
Whether or not “AGI era” turns out to be the right label, GPT-6 Astra is a real jump in benchmark performance and autonomy compared to what OpenAI shipped even a few months ago. The bigger question, one that will take longer than a launch event to answer, is whether that translates into reliable, safe results once millions of people start turning Astra loose on real work.
Note: Some rollout timing, regional availability, and benchmark figures are based on early reporting and OpenAI’s own launch materials. Exact numbers may be refined as OpenAI publishes its full model card.
