What happened when OpenAI temporarily paused new sign ups and upgrades for its $200 per month ChatGPT Pro plan on September 11, 2026, why did it take that step, what did OpenAI executives say about unprecedented demand for the newly launched GPT 6 Astra and the strain it placed o

OpenAI temporarily stopped new subscriptions and upgrades to its $200/month ChatGPT Pro (Pro 20X) tier on September 10–11 because the newly released GPT-6 Astra created exceptional demand and Pro users imposed the greatest load on its infrastructure. Existing $200 Pro members were not disrupted. 14

• What was paused—and what stayed open: New $200 Pro sign-ups and upgrades from Free, Go, Plus, and $100 Pro were paused. Existing $200 Pro subscriptions, and new or existing $100 Pro subscriptions, remained active. Astra also remained available through the API; the evidence indicates Go and Plus were unaffected. 13

• Reason and duration: Product leader Thibault “Tibo” Sottiaux said Pro usage placed the most strain on OpenAI’s systems, amid “unprecedented” Astra demand. Reports characterize the step as a capacity-protection measure while OpenAI added capacity, but the company had not publicly provided a firm end date for the pause. 47

• Astra launch and positioning: GPT-6 Astra began rolling out on September 3, 2026, as OpenAI’s frontier flagship and successor to GPT-5.6 Sol. Its standard API price was $10 per million input tokens and $50 per million output tokens. 58 Reports say OpenAI portrayed it as its most capable and aligned model; Greg Brockman reportedly called the moment the “AGI era,” while the company did not formally declare Astra to be AGI. 6

• Enterprise push: On September 10, OpenAI launched ChatGPT for Financial Services, combining Astra with built-in licensed data from LSEG, PitchBook, and Daloopa for investment bankers and equity researchers. 2 The product was developed with Morgan Stanley and Evercore, and was restricted to eligible institutions using ChatGPT Enterprise/Work access. 103

• Safety and cyber risk: Astra was reported as OpenAI’s first model to reach the “Critical” cybersecurity-capability threshold in its Preparedness Framework—triggering additional deployment restrictions. 8 Reports also say OpenAI slowed parts of frontier development after internal evaluations reached that threshold; however, the available evidence does not provide enough primary-source detail to specify every cyber capability or the exact duration and scope of the slowdown. 78

• Why interest was especially intense: Astra’s claimed performance was striking in agentic computer use, coding, mathematics, and cyber tasks—for example, reported OSWorld 2.0 performance of 72.6%, versus 65.7% for Sol. 12 But independent comparisons were more mixed: Artificial Analysis reportedly scored Astra at 61.2, near Sol’s 60.9 and below Anthropic’s Claude Fable 5.1 at 65.7. 7 Astra launched two days after Fable 5.1 at the same $10/$50 headline token price. 8

• Limits of the evidence: I found insufficient reliable evidence in the retrieved sources to substantiate the specific comparison with Grok 4.5 pricing, identify particular upcoming OpenAI events, or detail criticism from creatives. Those elements should not be treated as established without stronger primary reporting.