The Model in the Press Release Is Not the One You Get.

OpenAI released GPT-6 Astra yesterday. Greg Brockman said "Welcome to the AGI era," and on whether this particular model qualifies, "I think it might be about this model."

Here is the part nobody put in a headline. At least three different things are wearing that name right now, and the one being written about is not the one on your account.

Two disclosures before the argument. I pay OpenAI, Anthropic, xAI and Perplexity, so I am a customer of the company I am about to be skeptical of and of three of its competitors. And I sell implementation work, which means a post arguing that announcements are not deliveries is a post that sells what I do. Verify the rollout terms yourself before you take my word.

Three Products, One Name

Access started with a limited set of organizations in OpenAI's Daybreak Access program.

ChatGPT Plus, Pro, Business and Enterprise customers, plus API developers, get it "in the coming days."

The most capable cybersecurity functions are restricted to a small group of trusted testers, and it is not clear when or whether that widens.

So the model in the coverage, the model your subscription will eventually reach, and the model with the capabilities that generated the most alarm are three different products. Every article about the first one gets read as though it describes the third.

"In the Coming Days" Is Not a Date

I should be careful here, because a staged rollout is a defensible thing to do. If a model has capabilities that could cause harm at scale, releasing it gradually and holding back the sharpest parts is what a responsible vendor does. I would rather have that than the alternative.

But defensible for the vendor and usable for the buyer are different questions.

A roadmap needs a date. A budget needs a price. A security review needs a scope. "In the coming days" gives you none of those, and neither does a capability tier you cannot get into. If your Q4 plan has a line item that depends on a model announced yesterday, that line item is currently held together by a press release.

This is the same failure I keep seeing in mid-market technology plans. Somebody reads a launch, assumes availability, and builds a quarter around it. Then the thing arrives late, or arrives smaller, and the plan quietly rots.

The Evals Have No Numbers

OpenAI says Astra set new marks on biology, chemistry, medical and physics evaluations, and that it contributed to mathematical research on gaps between prime numbers.

I went looking for the numbers. Not one report I could find names a benchmark, gives a score, or offers a comparison to a prior model. I also could not reach OpenAI's own announcement page from where I was working, so everything I have described here comes from press coverage rather than the primary source, and you should weigh it accordingly.

That gap matters more than it sounds. "New marks on several evaluations" is not a claim you can put in a contract, hold a vendor to, or reproduce on your own workload. It is a direction, not a measurement.

If you are going to evaluate this model, the only number that will mean anything is the one you generate yourself, on your own task, with your own data.

What I'd Do This Week

Find every plan that assumes a model you cannot currently call. Not the ones that mention it. The ones that depend on it. Those are the ones that break.

Write down the tier you actually have access to. Not the tier in the announcement. If nobody on your team can name it, that is the finding.

Ask your vendors for dates in writing, in your contract, not in a blog post. A launch is marketing. A commitment is a term.

Build one eval on your own workload before the access arrives. When it does show up, you will be able to answer whether it helps you in a day instead of a quarter.

The Part That Matters

Astra may turn out to be everything OpenAI says. I have no way to know yet, and neither does anybody writing about it this week.

What I do know is that the gap between announced and available has become the most expensive assumption in enterprise technology planning, and it keeps getting wider while everyone watches the demos.

Plan against what you can call today. Everything else is a rumor with a version number.