OpenAI has a new flagship model, and per company tradition, it arrives with a name that sounds like a luxury sedan and benchmarks that sound made up. Meet GPT-6 Astra — the model OpenAI is calling its most capable and most aligned yet, which is corporate-speak for "please don't make us regret this."

The Numbers OpenAI Wants You to See

Astra rolled out in limited preview on September 3, 2026, before expanding to ChatGPT Plus, Pro, Business, and Enterprise users, plus the API, Azure, and AWS Bedrock. It scored 72.6% on OSWorld V2-Offline, a benchmark for operating real desktop software, up from 65.7% for the previous GPT-5.6 Sol — and it reportedly saturates FrontierMath Tier 4 at 98%, a benchmark specifically designed to be brutal.

Under the hood, Astra uses a "recurrent depth" architecture — looped transformers that let the model reason internally before answering, trained across more than 100,000 GPUs at OpenAI's Stargate facility in Texas. OpenAI is positioning it as state-of-the-art on computer use, browsing, software engineering, and cybersecurity work, not just chat.

Smarter Also Means Harder to Read

Here's the trade-off nobody put in the headline: the same recurrent-depth trick that makes Astra faster and sharper also makes its internal reasoning harder for humans to inspect. OpenAI calls it their most aligned model yet, but safety researchers have already flagged that a less legible thought process is a strange feature to ship alongside a model this capable of autonomous computer use.

For businesses, the real story isn't the benchmark chart — it's that a model this good at "using a computer" is inching toward doing the unglamorous middle-office work software has never quite automated: filling out forms, navigating internal tools, closing the loop on multi-step tasks nobody wanted to own.

Astra is less a chatbot upgrade and more a hint at what "software" stops meaning in a few years.

If you're wondering what a model like Astra could actually do inside your business — beyond the demo reel — that's the exact conversation James and the WTK team have with clients every week.

Source: The New Stack