GPT-6 Astra speeds up computer use and multistep office work; Enterprise access is off by default

OpenAI released GPT-6 Astra with computer use, browsing, coding, and multistep professional work as the stated focus. In OpenAI's own latency simulation on OSWorld 2.0 it reaches 72.6% at roughly 40 minutes per task, against 65.7% at roughly 75 minutes for GPT-5.6 Sol, and an updated Codex harness completes Mind2Web tasks 1.9x faster than the current Sol experience. On coding it scores 57.9% on Terminal-Bench 4.0 against Sol's 37.3%, and 63.9% on OpenAI's internal database-migration tasks against 42.7%. For long sessions, Codex can now keep notes across context windows instead of compacting them into one summary - an experimental setting in config.toml that OpenAI says becomes the default in the coming weeks. On security work the model is deliberately narrowed: OpenAI says it handles secure code review and patching but refuses more advanced tasks such as building proof-of-concept exploits. Why it matters: OpenAI calls Astra the world's most intelligent and aligned model, but the comparison table in the same post is narrower than the headline: on the third-party Artificial Analysis Intelligence Index v4.1.1 Astra scores 61.2, below Claude Fable 5.1 at 65.7 and Claude Opus 5 at 63.1. Where the reported gaps are large and specific is computer use, terminal and migration coding, and minutes spent per task. That makes this a per-workload decision rather than a blanket model switch, and the workloads it favors are exactly the long, tool-driven jobs where wall-clock time is the cost.

Try this: Run one real multistep job you already do - a browser research-and-write task, or a database migration - on gpt-6-astra and on your current model, and compare output quality and wall-clock time on the same input. Price both runs at Astra's $10 per million input and $50 per million output tokens (Fast mode is 2x the speed at 2x the price) before you move a route. Check availability first: rollout reaches ChatGPT Plus, Pro, Business, and Enterprise users plus the API, Azure, and Bedrock over the coming days, and on Enterprise an administrator has to enable it because access is off by default at launch.