Never lose a customer to a missed message
An AI agent trained on your own business, replying in seconds, in any language, on every channel your customers already use.
OpenAI began rolling out its newest flagship model on September 3. The company calls it the most capable system it has ever shipped. The GPT-6 Astra release arrived months later than originally planned. OpenAI said it added extra safeguards after a security incident earlier in the year, according to CNBC.
What the GPT-6 Astra release actually includes
Astra ships with a one-million-token context window. Its API price runs roughly 2.5 times higher than its predecessor, GPT-5.6 Sol. OpenAI set API access at $10 per million input tokens and $50 per million output tokens. Cached input, batch processing and a slower “flex” mode all cost less. The rollout started with a limited group of trusted partner organizations. A wider release to ChatGPT’s paid tiers, the API, and cloud partners including AWS follows in the coming days.
OpenAI’s own benchmark claims are striking. The company says Astra scores 98% on FrontierMath Tier 4. It also claims 99.9% on ARC-AGI-3 and 100% on ExploitBench, a security-focused benchmark. Those numbers come from OpenAI itself. They are not independently verified, and outside researchers have not yet had time to reproduce them at scale.
Why the release was delayed
OpenAI said it pushed back Astra’s launch after a breach linked to its Hugging Face-related infrastructure in July. The company used the extra time to add safeguards. It says those safeguards are meant to reduce the risk of serious misuse. Al Jazeera reported the launch is drawing heightened scrutiny. Astra’s advanced cybersecurity capabilities are a selling point, but they also raise questions about dual-use risk.
What OpenAI says about the safety tradeoffs
OpenAI said this week it believes the added safeguards “sufficiently minimize the risk of severe harm” ahead of a broader release. That is the company’s own assessment of its own product. It is not an independent audit. It is worth watching closely as more researchers and enterprise customers get hands-on access in the coming weeks.
What Astra’s rollout looks like from here
The next phase brings Astra to ChatGPT’s paid subscribers and general API access. At that point, far more developers and businesses can test its claimed capabilities against real workloads. That is a very different test than OpenAI’s own benchmark numbers. Enterprise security teams evaluating Astra’s cybersecurity features will likely publish independent findings first, given how directly those capabilities intersect with existing security tooling.
Competing labs are expected to respond within weeks with their own updated models, a pattern that has held throughout 2026 as each major release resets the competitive bar.
How rivals are likely to respond
OpenAI does not operate in a vacuum. Google, Anthropic and several well-funded startups have each shipped major model updates this year. Every time one lab leaps ahead on benchmarks, the others tend to respond within weeks rather than months. That pattern held throughout 2026, and there is little reason to expect it to break now.
What makes Astra different is the cybersecurity angle. Most prior releases competed mainly on reasoning and coding benchmarks. Astra’s marketed strength in exploit-related tasks puts security teams, not just developers, directly into the conversation about what a frontier model should be allowed to do.
Enterprise buyers evaluating Astra will likely ask pointed questions about access controls and misuse monitoring before deploying it widely. Vendors that sell security software are watching this rollout closely, since a genuinely capable exploit-finding model changes the calculus for both attackers and defenders.
A model marketed for its offensive security strength is exactly the kind of release that tends to draw regulatory attention in both the EU and UK, given each region’s growing focus on frontier AI oversight. How that scrutiny plays out here is likely to shape how future releases in this category get framed.
For now, most of what the public knows about Astra’s real-world risk profile comes from OpenAI’s own summary rather than an outside, independent review. That gap tends to narrow over the weeks after a major model launch, as more outside researchers get access and publish their own findings.
That review process, more than any single benchmark number, will ultimately decide how Astra gets remembered: as a genuine leap forward, or as a launch whose claims outran what the model could reliably deliver once wider testing began.
What people are asking
When did GPT-6 Astra launch?
OpenAI began rolling out GPT-6 Astra on September 3, 2026, starting with a limited preview for trusted partner organizations.
How much does GPT-6 Astra cost to use via the API?
OpenAI priced Astra at $10 per million input tokens and $50 per million output tokens, roughly 2.5 times the price of its predecessor, GPT-5.6 Sol.
Why was the release delayed?
OpenAI said it delayed the launch to add safeguards following a security incident linked to its Hugging Face-related infrastructure in July.
What is Astra’s context window?
Astra supports a one-million-token context window, according to OpenAI’s release details.
Have Astra’s benchmark scores been independently verified?
Not yet. The reported scores, including 98% on FrontierMath Tier 4 and 100% on ExploitBench, come from OpenAI’s own testing.
Sources
- CNBC — OpenAI announces rollout of GPT-6 Astra model. cnbc.com
- Al Jazeera — OpenAI unveils GPT-6 Astra amid rising scrutiny and safety concerns. aljazeera.com
Related coverage on Tamara News: the AI chatbot outage earlier this month and ChatGPT’s DSA designation in the EU.
Verification your users actually receive.
Send one-time passcodes over WhatsApp with a single API call. Replio can generate, hash and verify the code for you.