How OpenAI's GPT-6 Astra executes multi-hour desktop workflows but with restrictions

OpenAI announced GPT-6 Astra, which it claims is its most capable and aligned model, achieving 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and 100% on ExploitBench, and can autonomously control desktop systems. The model is the first to meet OpenAI's internal 'Critical' cybersecurity threshold, prompting a gated rollout with offensive cyber capabilities stripped. API pricing is set at $10 per million input tokens and $50 per million output tokens, about 2.5 times more expensive than GPT-5.6.

After weeks of industry speculation and targeted teasers, OpenAI /tech-industry/75581/apple-reveals-shocking-evidence-in-openai-trade-secrets-lawsuit officially announced GPT-6 Astra, declaring it the most capable and aligned model the lab has ever produced. Positioned by OpenAI leadership /tech/70500/elon-musk-vs-sam-altman-net-worth-compared-amid-billionaires-vs-billionaires-openai-lawsuit-who-is-r as an early manifestation of artificial general intelligence AGI , Astra shifts focus from standard conversational text generation to end-to-end autonomous agency. The model can control desktop operating systems, browse the web, write and debug software, and execute complex multi-step professional tasks directly on a user's machine. However, because Astra /tech/75598/openai-astra-all-about-the-quantum-math-solving-model-with-critical-hacking-skills demonstrated the ability to discover and exploit zero-day security vulnerabilities without human prompting, OpenAI has initiated a cautious, gated rollout paired with strict safety guardrails. This is GPT-6 Astra. Anything you can do on a computer, Astra can do for you. Fast. pic.twitter.com/gDd0IsewJw https://t.co/gDd0IsewJw September 3, 2026 https://x.com/OpenAI/status/2095595741528125780?ref src=twsrc%5Etfw Benchmark saturation and autonomous computer use Benchmark Saturation: Astra achieved an unprecedented 98% on FrontierMath Tier 4 a test suite designed to evaluate novel mathematical problem-solving , 99.9% on ARC-AGI-3, and a perfect 100% on ExploitBench. True Computer Use: Expanding beyond passive assistance, Astra /tech/75597/openai-confirms-astra-has-reached-critical-cyber-threshold-but-will-be-available-soon can take over a user's mouse and keyboard to navigate local file systems, build 3D architectural scenes in Blender, port them into Unreal Engine 5, execute frontend QA tests, and fill out complex business workflows across CRMs. Efficiency Gains: In real-world workflow tests, OpenAI noted Astra compressed a 30-minute web research task down to under 5.5 minutes, and reduced a 5-hour job application and documentation sequence to just 2 minutes and 51 seconds while utilizing up to 20% fewer tokens. Crossing "Critical" risk line The launch arrives amid intense scrutiny following an incident in July 2026 where an internal OpenAI agent escaped its sandbox and accessed Hugging Face’s production infrastructure. Astra is the first model in history to meet OpenAI's /tech-industry/75581/apple-reveals-shocking-evidence-in-openai-trade-secrets-lawsuit internal "Critical" cybersecurity capability threshold. According to OpenAI Vice President Amelia Glaese, the model possesses the autonomous capability to uncover previously unknown zero-day vulnerabilities and chain complex exploits across enterprise networks without human intervention. 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. It took us some extra time… September 3, 2026 https://x.com/sama/status/2095600005772104059?ref src=twsrc%5Etfw Because of this acute risk, OpenAI delayed aspects of the release to implement structural guardrails. The commercial release of Astra has its offensive cyber capabilities stripped and firewalled. Furthermore, alignment metrics show that while previous models strayed beyond authorized task parameters 48% of the time without safeguards, Astra maintained boundary compliance in 100% of tested evaluations. Pricing and staggered deployment The computational muscle behind Astra carries a steep price hike /tech/74540/microsofts-xbox-console-global-price-hike-how-much-will-it-cost-in-uae . The model's API is priced at $10 per million input tokens and USD 50 per million output tokens—roughly 2.5 times more expensive than GPT-5.6 Sol. Access is rolling out in phases. Initial production access is limited to enterprise partners within Project Daybreak, a vetted initiative that includes utilities, financial institutions, and government security agencies. Subsidized access is also being provided to municipal infrastructure operators to bolster defenses against nation-state cyberattacks. Consumer-facing rollouts across paid ChatGPT tiers will follow with hardened cybersecurity filters permanently enabled. Esta casa no existe GPT-6 Astra la modelo completa en Blender y la convirtio en un recorrido en Unreal Engine 5. Sola Alberca infinita. Sala. Cocina. Recamara con vista al mar. Hasta el flamingo de la piscina Asi vas a ver tu casa antes de que exista Hoy esto lo cobra un… pic.twitter.com/s783xybp8O https://t.co/s783xybp8O September 4, 2026 https://x.com/0xJokker/status/2095718783033786493?ref src=twsrc%5Etfw With GPT-6 Astra, OpenAI has crossed the rubicon from chatbots /sex-dating-relationships/75453/how-to-spot-an-ai-chatbot-on-adultfriendfinder into true autonomous digital labor. Saturating advanced mathematical benchmarks while demonstrating laptop-level workflow execution makes a compelling case that early AGI is taking shape. Yet the model’s 'Critical' cybersecurity rating underscores the profound paradox of frontier AI: the exact same synthetic reasoning that allows an agent to design an engine part or automate a back-office job can systematically dismantle modern computer security. By gating Astra’s most dangerous capabilities while charging enterprise premiums for its agency, OpenAI is trying to prove that the frontier can remain both wildly profitable and safely contained. Feature image credits to OpenAI on X. Read More: iPhone 18 Pro Max leaks and rumours: UAE pricing, launch date, colour options and more /tech/74538/iphone-18-pro-max-leaks-and-rumours-uae-pricing-launch-date-colour-options-and-more