OpenAI cuts GPT-6 Astra usage limits up to 4x, warns of Pro signup pause

Only days after GPT-6 Astra launched and following a September 5 banked-usage reset, OpenAI reportedly cut usage limits by up to 4x for heavy Plus, Pro, and Business users, according to reporting aggregated by explainX and others. A September 10 update goes further, signaling OpenAI may pause new Pro subscriptions altogether if Astra demand keeps climbing — a striking admission for a company simultaneously touting the model's GA across Bedrock and Microsoft Foundry.

The root cause is capacity. Astra is a heavier, more compute-intensive model, and demand outstripped supply almost immediately. The situation echoes prior launch crunches: Sam Altman reportedly apologized after Pro and Plus users couldn't access the new model, and the backdrop is OpenAI's roughly $750B in planned compute spend through 2030 — a number that itself signals how supply-constrained the frontier has become.

The community reaction is sharp. On X and Reddit, one user reported burning through a $200/month Codex plan in 12 hours; OpenAI's Tibo Sottiaux included a Rick Astley reference in a reset note that many read as deflecting a genuine capacity problem. The episode crystallizes a theme running through this week's news — model-capability announcements are increasingly gated by raw compute, and 'available' now often means 'available, rate-limited.' Watch whether OpenAI actually pauses Pro signups, how the limits interact with enterprise Astra on Bedrock/Foundry, and whether cheaper rivals like DeepSeek and Grok capture frustrated power users.