OpenAI announced GPT-6 Astra, which it describes as its most capable and most aligned model, now available in ChatGPT Work, Codex, and the API. According to OpenAI, Astra can write code and operate within applications people already use, including those without an API, so businesses can deploy it without extensive preparation or custom integration. The company reports early customer uses ranging from optimizing GPUs to spotting discrepancies in financial statements and producing on-brand decks. OpenAI states its own engineering team used Astra to resolve a memory-allocation bottleneck, achieving 25x lower turn latency with roughly 30% higher peak memory use. Pricing starts at $10 per million input tokens and $50 per million output tokens. On OpenAI's internal computer-use safety benchmark, the company says Astra produced unintended outcomes 89% less often than GPT-5.6 Sol and 74.7% less often than Claude Fable 5.1. New enterprise admin controls allow organizations to restrict access to approved websites and applications, manage uploads and downloads, and require approval before consequential actions. OpenAI also states Astra is the first model to reach the Critical cybersecurity capability threshold under its Preparedness Framework. Enterprise access is off by default at launch. These figures come from OpenAI's own announcement and have not been independently verified.
- Pricing starts at $10 per million input tokens and $50 per million output tokens
- Available in ChatGPT Work, Codex, and the API; enterprise access off by default at launch
- OpenAI claims 89% fewer unintended outcomes than GPT-5.6 Sol on its internal safety benchmark
- First model OpenAI says reaches the Critical cybersecurity threshold under its Preparedness Framework
What it means for you
OpenAI has released a newer model that it says can work directly inside the software your team already uses, rather than needing a developer to wire things together first. The headline claims — speed, cost efficiency, fewer mistakes — all come from OpenAI's own testing, so treat them as marketing until independent reviewers weigh in. The genuinely useful change for smaller teams is the new admin controls that let you require approval before the AI takes any consequential action.
Try this
If you use ChatGPT Business or Enterprise, ask your account admin whether Astra is enabled and, before turning it loose on real systems, set up the confirmation policies so a human approves anything that deletes, sends, or shares data.
Who should care
Small businesses already paying for ChatGPT Work or Codex, and teams weighing letting AI operate directly inside their existing apps.
Skip this if
You're using free or personal ChatGPT for writing and research — this is an enterprise-tier release and changes nothing about your day-to-day use.
Sources: OpenAI — read the original