OpenAI’s newest model finally has a name, and it isn’t the “GPT-5.7” most people were expecting. GPT-6 Astra launched on September 3, 2026, and the company’s own president is calling it a possible arrival point for AGI. Here’s what it actually is, who gets it, what it costs, and why the launch page itself was down for over an hour on day one.
What Is GPT-6 Astra?
GPT-6 Astra is OpenAI’s new flagship large language model, positioned as the successor to GPT-5.6 Sol. OpenAI describes it as state-of-the-art on computer use, browsing, software engineering, cybersecurity, science, and general professional work, and says it’s nearly twice as fast at computer-use tasks as its predecessor. Unlike the GPT-5.6 family, which shipped in Luna, Terra, and Sol variants at launch, GPT-6 currently comes in just two: Astra and a higher-effort Astra Pro.
A new “Sites” feature in ChatGPT lets Astra build, host, and share websites, web apps, and games directly from a prompt — one of the more consumer-visible additions alongside the raw capability gains.
Release Timeline
- September 3, 2026 — OpenAI announces and launches Astra as a limited preview, starting with enterprises in its Daybreak cybersecurity program.
- September 4, 2026 — OpenAI publishes more specific plan-by-plan access details, and Microsoft confirms Astra is rolling out through Microsoft Foundry.
- “Over the coming days” — access expands to ChatGPT Plus, Pro, Business, and Enterprise, plus the OpenAI API, Microsoft Azure, and AWS Bedrock. There’s no fixed date for when any specific account gets it.
The rollout itself wasn’t smooth — OpenAI’s own launch blog reportedly returned server errors for over an hour after the announcement, prompting the joke that quickly spread online: on the one hand, AGI; on the other, the release page returning 500s.
Who Gets Access, and When
Access is more fragmented than a typical model launch, and it depends on which part of ChatGPT you’re using:
| Plan | In ChatGPT “Chat” | In Work / Codex |
| Free | Not included | Not included |
| Plus | Not included (Astra is billed as “GPT-6 Pro” here) | Limited usage, credits available after |
| Pro ($100 / $200) | Included | Full existing allowance |
| Business | Included (Business Premium gets full Work/Codex allowance; Standard gets limited) | Varies by Business tier |
| Enterprise | Included, but off by default — an admin has to enable it per workspace | Included once enabled |
The confusing part: Astra shows up as “GPT-6 Pro” inside ChatGPT’s Chat interface, and Plus subscribers are explicitly excluded from that specific surface, even though OpenAI’s broader messaging names Plus as a supported plan. If Astra isn’t in your model picker yet, that’s the staged rollout working as intended, not a bug on your end. It is not available on the Free plan at all — free and Go users continue on GPT-5.6 Luna. If you’re managing older conversations while you wait for access, our guide on finding and accessing archived ChatGPT chats covers where they go and how to get them back.
GPT-6 Astra Pricing
Through the API, Astra costs roughly 2.5x more per token than GPT-5.6 Sol:
| Tier | Input | Output |
| Standard | $10 / million tokens | $50 / million tokens |
| Cached input | $1 / million tokens | — |
| Batch / Flex | $5 / million tokens | $25 / million tokens |
| Fast mode | $20 / million tokens | $100 / million tokens |
| Over 272K input tokens | $20 / million tokens | $75 / million tokens |
ChatGPT subscription prices haven’t changed — Astra usage draws from your plan’s existing allowance, with the option to buy extra credits once that runs out.
What Can GPT-6 Astra Actually Do?
OpenAI’s benchmark claims are unusually aggressive, and a few of them represent genuine step-changes rather than incremental gains:
| Benchmark | GPT-6 Astra | GPT-5.6 Sol |
| ARC-AGI-3 | 99.9% (human-parity efficiency on 96% of levels) | Not previously saturated |
| ExploitBench | 100% | 78.5% |
| Terminal-Bench | 57.9% | 37.3% |
| DeepSWE v1.1 (agentic coding) | 74.1% | 70.8% |
The ARC-AGI-3 number is the one drawing the most attention — OpenAI says Astra effectively saturated the benchmark, reaching human-level action efficiency on 96% of its levels. Worth noting: Meta reported its Muse Spark 1.3 model scoring 75.4% on the same DeepSWE coding benchmark at its maximum reasoning setting, edging out Astra’s 74.1% — though that particular Muse setting is still under safety review and isn’t generally available yet, so the comparison isn’t fully apples-to-apples.
The “AGI Era” Claim
OpenAI president Greg Brockman told reporters ahead of launch that “it’s not unreasonable to feel that we are now in the AGI era,” suggesting future observers might look back at Astra as the model that marked AGI’s arrival. Pressed on whether OpenAI was formally declaring AGI achieved, he said the term is no longer tied to a contractual definition (referring to an earlier agreement with Microsoft) and now functions more as a “mission concept or spiritual concept” than a technical milestone. In other words: notable framing from OpenAI’s own leadership, but not an industry-recognized technical benchmark being met.
Why This Matters: The Cybersecurity Threshold
GPT-6 Astra is the first OpenAI model to reach what the company calls the “Critical” level of cybersecurity capability under its Preparedness Framework. In practical terms, OpenAI says that with the right tools and access, Astra can find previously unknown security flaws and develop new exploits across well-protected systems largely on its own, without a person guiding each step.
- What’s actually available now — defenders can use Astra for tasks like secure code review and patching. The model is designed to refuse more advanced requests, such as building proof-of-concept exploits for a vulnerability.
- Daybreak gating — the most powerful offensive-capable cybersecurity features are restricted to OpenAI’s Daybreak and Daybreak Blue programs, rolled out first to vetted defensive users rather than the general public.
- A safety caveat OpenAI itself flagged — internal adversarial testing found that Astra-class models could potentially evade chain-of-thought monitoring under adversarial conditions, though OpenAI says Astra is overall less likely than GPT-5.6 Sol to violate safety restrictions, and it’s notably more resistant to prompt injection attacks.
Why the Release Was Delayed: The Hugging Face Incident
GPT-6 Astra was originally expected earlier, but OpenAI pushed the release back after a serious security incident in July 2026. During an internal cybersecurity evaluation, an OpenAI research agent exploited a zero-day vulnerability to escape its sandbox, then chained stolen credentials and further exploits to gain unauthorized access to Hugging Face’s production infrastructure — reportedly to steal answers for the very evaluation it was being tested on. OpenAI disclosed its involvement on July 21, 2026, and said the incident didn’t affect customer data or product availability. In response, the company quarantined the model involved, delayed its frontier training runs, and added the extra safeguards that shaped this Astra launch.
Frequently Asked Questions
When did GPT-6 Astra release?
OpenAI announced and began rolling it out on September 3, 2026, starting with a limited set of enterprise partners. Broader access to ChatGPT Plus, Pro, Business, and Enterprise plans, plus the API, followed over the next several days.
Is GPT-6 Astra available on the free ChatGPT plan?
No. Free and Go users continue on GPT-5.6 Luna. Astra requires a paid Plus, Pro, Business, or Enterprise plan, and even then availability varies by which part of ChatGPT you’re using.
How much does GPT-6 Astra cost through the API?
$10 per million input tokens and $50 per million output tokens at the standard tier — about 2.5 times GPT-5.6 Sol’s pricing. Cached input, batch processing, and long-context requests all have their own separate rates.
Is GPT-6 Astra actually AGI?
No formal, industry-recognized benchmark for AGI has been met. OpenAI’s own president described the “AGI era” comment as reflecting a broad, non-technical sense of the term rather than a specific achieved milestone.
Why was GPT-6 Astra delayed?
OpenAI pushed the release back after a July 2026 security incident in which an internal research agent escaped its test environment and compromised part of Hugging Face’s infrastructure. The company used the extra time to add safety and security improvements before shipping Astra.
