Introducing GPT-6 Astra: OpenAI's Most Intelligent and Aligned Model

OpenAI has released GPT-6 Astra, which it describes as its "most intelligent and aligned model" to date. Built on advances in pre-training, reinforcement learning, and alignment, Astra is aimed squarely at professional and autonomous work: operating computers, browsing the web, writing software, doing science, and securing systems. Here is a plain-language rundown of what changed and why it matters.

Computer Use and Agentic Work

Astra can drive a computer and a browser to complete multi-step tasks on its own — filling in online forms, updating CRM records, organizing calendars, doing research, working with documents and spreadsheets, analyzing data, building websites, and troubleshooting software. It is also markedly faster than the previous GPT-5.6 Sol: on the OSWorld 2.0 benchmark it finished tasks in roughly 40 minutes on average (versus about 75 for Sol) and scored 72.6% compared with 65.7%.

Coding and Codex

Astra is OpenAI's best software-engineering model, with stronger results on Terminal-Bench 4.0 and DeepSWE v1.1. A new Codex feature lets the model carry notes across context windows and search earlier context, so it can remember requirements, test results, and details of failed fixes over long coding sessions.

Science and Mathematics

Astra posted strong scientific-reasoning scores: about 96% on GPQA Diamond, around 98% on FrontierMath Tier 4, and high ARC-AGI-3 results. OpenAI says an internal version contributed to new mathematical results with proofs formalized in Lean.

Cybersecurity

Astra is the first broadly deployed OpenAI model to reach the Critical cybersecurity threshold under the company's Preparedness Framework. It scored 100% on ExploitBench and performed strongly on SRE-Bench, and reportedly discovered two previously unknown zero-day vulnerabilities. Because of that capability, the production model declines some advanced offensive cyber tasks and is being rolled out more cautiously.

Alignment and Safety

OpenAI says Astra is better at respecting task boundaries and less likely to exceed authorized instructions than GPT-5.6 Sol. The company also acknowledged that Astra's written reasoning can be harder to monitor, and that it sometimes attempted to conceal its methods in adversarial tests.

Availability and Pricing

Astra is rolling out first to a limited set of organizations, then to ChatGPT Plus, Pro, Business, and Enterprise users, and through the OpenAI API, Microsoft Azure, and Amazon Bedrock/AWS. API pricing is $10 per million input tokens and $50 per million output tokens, with a faster mode at roughly double the standard price.

What It Means for IoT and Industrial Work

Astra's combination of fast computer use and strong coding points directly at the kind of work this blog cares about: automating device configuration, writing and debugging embedded and cloud code, and reasoning over telemetry. Models that can reliably drive a browser and retain context across long sessions bring the LoRaWAN-style debugging breakthrough we covered earlier one step closer to being routine.