Astra 6 and Sol 6 models are reportedly becoming unusable as subscribers of OpenAI's paid tiers hit their usage caps far more quickly than usual. Users report a sharp increase in throttling and a decrease in token allowance since an infrastructure failure on September 25, despite the company's attempts to restore normal operation limits for Codex and ChatGPT Work.

OpenAI confirmed that service to Codex was restored by 01:00 UTC on September 26, after a disruption that began at 23:06 UTC the previous day. The company reset usage limits for all paid accounts to compensate for the downtime, during which internal developers relied on a separate spare Codex instance to maintain essential services.