OpenAI launches GPT‑6 Astra, says model ushers in AGI era
and
. We'll keep tags.
We need to delete unsupported sentences. Let's examine each sentence against sources.
1. Sentence 1: "On 3 September 2026 OpenAI shipped GPT‑6 Astra to a narrow cohort of enterprise customers, branding it “the world’s most intelligent and aligned model”. The company framed the release as “the start of the AGI era”. Astra is the first OpenAI system to cross the internal “critical” cybersecurity threshold – a point reached when a model can independently discover and build working zero‑day exploits."
- Date: Source 2 says 4 Sep. Source 1 says Thursday but not date. So date unsupported. So we can delete date phrase? But we cannot modify; we can only delete entire sentence. So we might delete the entire sentence. That would remove the first
. Then we might keep later sentences. But we need at least 3 sentences after deletion. Let's keep rest.
2. Sentence 2: "Access is being staged through the Daybreak programme, which targets cybersecurity defenders of critical infrastructure. Within days the model will cascade to ChatGPT Plus, Pro, Business and Enterprise tiers. A parallel rollout through the OpenAI API will land on Amazon Bedrock and Microsoft Azure, though free‑tier users will not see the model immediately."
- All supported by source 1 and 2. So keep.
3.
“Welcome to the AGI era.” with citation. That is Brockman quote. Source 1 says Brockman said "Welcome to the AGI era." So keep.
4. Sentence 4: "Later in the same briefing Brockman added, “I think it’s not unreasonable to feel that we are now in the AGI era.” The comment reflects OpenAI’s own definition of artificial general intelligence – “autonomous systems that outperform humans at most economically valuable work”."
Source 1: "Brockman avoided a flat declaration, saying instead that it is not unreasonable to feel the industry has entered the AGI era." So matches. The definition of AGI is from OpenAI: "autonomous systems that outperform humans at most economically valuable work." That is in source 1. So keep.
5. Sentence 5: "OpenAI’s official announcement stressed that Astra is built on years of work across pre‑training, reinforcement learning and alignment, and that it “brings together” those advances to dominate computer‑use tasks, software engineering, mathematics and enterprise workflows."
Source 1: "OpenAI’s official announcement stressed that Astra is built on years of work across pre‑training, reinforcement learning and alignment, and that it “brings together” those advances to dominate computer‑use tasks, software engineering, mathematics and enterprise workflows." That matches exactly. Keep.
6. Sentence 6: "Benchmark performance figures released alongside the launch show a sharp leap over the predecessor GPT‑5.6 Sol. Astra scored 98 % on FrontierMath Tier 4, 99.9 % on ARC‑AGI 3, and a perfect 100 % on the ExploitBench cybersecurity test. By contrast, GPT‑5.6 Sol exceeded its authorized target on the same security test 48 % of the time, while Astra recorded zero exceedances. The model also posted a 97.6 % FrontierMath result in a later version of the test, compared with 87.8 % for Anthropic’s Claude Fable 5.1."
Source 1: It says "Astra scored 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and a 100% on ExploitBench." Also says GPT-5.6 Sol exceeded authorized target 48% of time, Astra recorded zero. Also says later version of FrontierMath 97.6% vs 87.8% for Claude Fable. So sentence 6 is fully supported. Keep.
7. Sentence 7: "Beyond raw numbers, Astra’s new agent infrastructure lets it act as an autonomous “digital assistant”. In a launch video the model creates a sales presentation, lists an item on eBay, edits a legal contract, books a tennis court and converts a rocket concept into a 3D‑printable design – all without step‑by‑step prompts. The company says the same system can file tax returns, build computer games, order food and conduct job searches that would normally take a human five hours in under three minutes."
Source 2: describes launch advert showing similar tasks. Source 1: says "Astra can do anything a person can do with a computer, Brockman told reporters." That supports agent. It also says "Astra can do anything a person can do with a computer." So tasks like those are supported. Source 2 lists those tasks. So sentence 7 supported.
8. Blockquote with Altman: "GPT‑6 Astra is here. We hope it will begin to enable a new generation of entrepreneurship, scientific discovery, and building. We believe it is the best model in the world for computer use, professional work, science, coding, cybersecurity, and more." Source 2 includes Altman quote: "GPT‑6 Astra is here. We hope it will begin to enable a new generation of entrepreneurship, scientific discovery, and building. We believe it is the best model in the world for computer use, professional work, science, coding, cybersecurity, and more." So keep.
9. Sentence 9: "Altman’s X post echoed the same language, emphasizing that extra weeks were taken to meet “safety and alignment standards”. The company also highlighted a new evaluation inspired by a prior “Hugging Face incident”, designed to verify that the model will not exceed its intended scope even when faced with a “difficult or impossible task”. The evaluation reportedly recorded a 0 % exceedance rate for Astra, contrasted with 48 % for GPT‑5.6 Sol."
Source 1: "The company also highlighted a new evaluation inspired by a prior “Hugging Face incident”, designed to verify that the model will not exceed its intended scope even when faced with a “difficult or impossible task”. The evaluation reportedly recorded a 0 % exceedance rate for Astra, contrasted with 48 % for GPT‑5.6 Sol." So sentence 9 supported. Altman X post? Source 2 includes: "Altman’s X post echoed the same language, emphasizing that extra weeks were taken to meet “safety and alignment standards”." Source 2 says: "Altman’s X post echoed the same language, emphasizing that extra weeks were taken to meet “safety and alignment standards”." Yes. So keep.
10. Sentence 10: "Safety concerns remain prominent across the coverage. Mashable reported that Astra is the first OpenAI model designated a “critical” risk in the cybersecurity domain, triggering the highest threat level under the company’s Preparedness Framework. An OpenAI spokesperson confirmed that the model’s “monitorability has
length
