OpenAI just paused its own Pro plan because Astra broke it
Issue #035 · 4 min read
OpenAI just paused its own Pro plan because Astra broke it
Demand outran capacity. Plus new Agents API, GitLab's Astra numbers, and Meta's Muse problem.
The big story
OpenAI has stopped selling its own $200-a-month Pro plan.
Not because of a bug. Because too many people wanted it.
The company says Pro subscriptions put the most strain on its systems, so it's pausing new sign-ups while it adds capacity — that's TechCrunch. The strain traces back to Astra, the new model behind the plan.
There's a second data point sitting right next to this one. The $2,000-a-month Pro 20x tier — one rung up from regular Pro — is now unavailable for purchase too, according to a screenshot circulating on X.
Read those two facts together and a pattern shows up. OpenAI shipped a model good enough that its own top-paying customers are trying to buy more of it than the company can serve.
That's a nice problem in theory. In practice it means every enterprise buyer currently negotiating a Pro or Astra contract just lost their easiest point of comparison — the sign-up page.
If you're mid-negotiation with OpenAI right now, ask your rep directly whether capacity constraints affect your renewal terms. Silence on a pricing page is not the same as a "no."
What shipped
OpenAI opened up an Agents API for building and running cloud agents, powered by the Codex harness, with support for long-running sessions and tool use — full details here.
This is the plumbing behind a lot of what OpenAI has been announcing lately. If your team has been hand-rolling agent orchestration on top of the Assistants API, this is worth an afternoon to evaluate before you build more on the old foundation.
GitLab put numbers on what switching to GPT-6 Astra actually does to a CI run. In GitLab's own internal evaluation, Astra finished a typical run 43.4% faster than the previous model and used 42.7% fewer tokens, while completing every task in the benchmark — GitLab's writeup.
That's a rare thing: a vendor publishing a real before-and-after number instead of "faster and cheaper." If you run agentic CI tasks — dependency bumps, build fixes — and you're paying per token, this is a concrete lever to pull this quarter.
Meta's Muse is now the No. 2 app in the US, per TechCrunch — but a hands-on from The Verge says it "works and creeps me out." Fast growth and a genuinely uneasy first impression are not mutually exclusive, and if you're tracking Muse as a competitive signal, read both pieces before you draw a conclusion from the ranking alone.
What I'd actually do this week
Check whether your org's OpenAI seats are Pro or Pro 20x, and confirm your renewal date isn't sitting inside this capacity crunch — ask your account rep plainly.
If your CI pipeline runs agentic jobs on an older model, pull your own token and runtime numbers before deciding whether GitLab's 43%/42.7% figures would hold for your workload.
Skim the Agents API docs even if you're not building agents yet — it's the base a lot of vendor integrations will sit on by next quarter.
Reply if you've hit OpenAI's new sign-up pause yourself — I want to know what your account rep is telling paying customers.