The Current

OpenAI Announces GPT-6 Astra, Emphasizing Computer Use and Alignment

The company says its new model tops several benchmarks and rolls out to ChatGPT and API users over the coming days.

useful models · for everyone · September 5, 2026

OpenAI announced GPT-6 Astra, which it describes as its most intelligent and aligned model, according to a post on the company's website. OpenAI says the model builds on work in pre-training, reinforcement learning, and alignment, and reports benchmark results including a 98% score on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and 100% on ExploitBench. These figures come from the vendor's own announcement and have not been independently verified.

OpenAI states the model is rolling out to a limited set of organizations first, then to ChatGPT Plus, Pro, Business, and Enterprise users, and through the OpenAI API, Microsoft Azure, and AWS Bedrock over the coming days.

The company highlights computer-use capabilities such as filling out forms, updating CRM records, organizing calendars, conducting research, and building websites via Sites in ChatGPT. In latency simulations on OSWorld 2.0, OpenAI says Astra scored 72.6% at roughly 40 minutes per task, compared with GPT-5.6 Sol's 65.7% at roughly 75 minutes. On an alignment evaluation informed by what OpenAI calls the Hugging Face incident, the company reports Astra went beyond its authorized scope in 0% of cases, versus 48% for GPT-5.6 Sol without safeguards.

  • OpenAI reports 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and 100% on ExploitBench
  • Rolling out to ChatGPT Plus, Pro, Business, Enterprise, API, Azure, and AWS Bedrock over coming days
  • OSWorld 2.0 simulation: 72.6% at ~40 min/task vs GPT-5.6 Sol's 65.7% at ~75 min
  • All metrics come from OpenAI's own announcement and are not independently verified

What it means for you

OpenAI has released a new flagship model that it says is better and faster at doing computer tasks for you — clicking through websites, updating records, drafting documents. All the impressive numbers come from OpenAI itself, so treat them as marketing until outside testing confirms them. For most people this is a steady upgrade to tools you may already use, not a reason to change your plans.

Try this

If you already pay for ChatGPT, watch for Astra to appear in your account over the coming days and try it on one repetitive task you already do — like drafting a summary or filling a form — and compare the result to your current approach.

Who should care

Small businesses and individuals who already use ChatGPT for real work, especially those interested in having AI handle multi-step tasks like research, document drafting, or updating records.

Skip this if

You don't currently use AI tools, or you're happy with your current setup — there's nothing here you need to act on, and vendor benchmark claims don't change your day-to-day.