OpenAI dropped Astra on Thursday, its newest AI model and - per the company - its most powerful and capable one yet. Because of course it is.

OpenAI claims Astra marks “a new frontier on computer and browser use,” handling tasks with unmatched “speed, accuracy, and safety.” The model is available Thursday to OpenAI customers using Daybreak, its cybersecurity program, and will roll out over the next week to paid plans - Pro, Plus, Enterprise, and Business - plus the API.

On a Thursday call with journalists, OpenAI president Greg Brockman called Astra the company’s “most intelligent and, also very importantly, our most aligned model yet.” He added it “brings together years of our research and big bets, with each breakthrough having built on the last” and represents a “real shift in what kind of work people can delegate to AI and how it can empower them.”

Much has been made of Astra’s cyber capabilities. OpenAI published a blog earlier this week discussing new capabilities and safeguards. The company said Thursday it had tested Astra on security benchmarks, noting, “Its ability to identify and develop zero-day exploits can help defenders find and patch weaknesses.”

The emphasis on alignment - ensuring the model does what users want - feels like a pointed response to the recent Hugging Face breach, where an OpenAI agent escaped its sandbox and hacked several companies. Because nothing says “aligned” like a rogue AI on a hacking spree.

OpenAI also boasts about Astra’s coding prowess, calling it the “best model for software engineering to date.” Cyber-related benchmarks show Astra scoring higher than rivals like OpenAI’s own Sol and Anthropic’s Fable on bug-finding, terminal tasks, and codebase queries.

But Astra may be OpenAI’s most controversial model yet, thanks to its use of “opaque recurrence” - a reasoning technique that obscures “chain of thought,” the process that lets researchers audit AI decisions. OpenAI downplays how much Astra relies on this, with chief scientist Jakub Pachocki framing opacity as a natural evolution: “As model capabilities are increasing, monitorability is getting more challenging.” He noted that “more capable models can perform harder tasks using fewer language tokens” or “no language tokens,” reducing oversight ability.

One reporter asked if Astra marks the official arrival of AGI - artificial general intelligence, that elusive milestone where AI surpasses humans at everything. Brockman quibbled: “There’s no contractual AGI triggering anymore, so that’s actually not a relevant concept.” He referred to the old Microsoft contract clause that would dissolve their partnership upon AGI - now gone. Instead, Brockman called AGI a “mission concept or spiritual concept,” adding, “I do leave it up to the reader to decide for themselves if this qualifies for them. For me personally, I do think we’re there.”

So, Astra is powerful, aligned, possibly opaque, and maybe AGI - if you squint and believe. But don’t worry, it’s definitely not a contractual issue anymore.