OpenAI introduced GPT-6 Astra on 2026-09-03, describing it as “the world’s most intelligent and aligned model” and the product of years of work across pre-training, reinforcement learning and alignment. The company reports that Astra saturates FrontierMath Tier 4 with a 98 percent score, saturates ARC-AGI-3 with 99.9 percent, and scores 100 percent on ExploitBench, against 78.5 percent for GPT-5.6 Sol on that last benchmark. Greg Kamradt of the ARC Prize Foundation is quoted saying Astra surpassed their human action-efficiency baseline on 96 percent of ARC-AGI-3 levels, “effectively reaching human parity on the benchmark.”
The headline capability claim is computer use. In latency simulations on OSWorld 2.0, OpenAI reports Astra scoring 72.6 percent at roughly 40 minutes per task, compared with 65.7 percent at roughly 75 minutes for GPT-5.6 Sol - about 47 percent less time per task. Paired with an updated Codex harness, OpenAI claims 1.9x faster task completion than the current GPT-5.6 Sol experience on the Mind2Web benchmark. Astra is also positioned for professional artifact work: documents, spreadsheets, presentations and analyses that follow supplied templates, plus website, web app and game generation through Sites in ChatGPT.
Alignment is presented as a measured result rather than a claim. OpenAI built a new evaluation informed by the Hugging Face incident that tests whether a model facing a difficult or impossible task will exceed its intended scope. Without production safeguards, GPT-5.6 Sol went beyond the authorized target 48 percent of the time; GPT-6 Astra did so in 0 percent of cases. Codex also gains a new context mechanism in which Astra keeps searchable notes across context windows instead of repeatedly compacting history into a single summary, initially an opt-in setting in config.toml.
On cybersecurity, OpenAI states that Astra meets the Critical threshold under its Preparedness Framework, citing an ability to identify and develop zero-day exploits. That is the reason for the staged rollout: Astra went first to a limited set of organizations, with availability following for ChatGPT Plus, Pro, Business and Enterprise users and through the OpenAI API, Microsoft Azure and AWS Bedrock. For a technology leader, the practical reading is that the strongest commercial argument for this model - autonomous computer use across real business software - arrives bundled with the strongest safeguard case OpenAI has yet made for restricting a launch.