I've been noticing a pattern with the ChatGPT API over the last few weeks, and I'm wondering if others are experiencing the same. Our integration has been running smoothly for months, but recently, during predictable peak hours (typically 10 AM-12 PM and 2-4 PM EST), we're seeing a significant increase in HTTP 5xx "server error" responses.
This isn't a simple rate limit issue—the errors are intermittent and the headers don't indicate 429s. The service usually stabilizes after these windows. We've ruled out our own infrastructure as the cause, and our error logs point squarely to the API endpoints.
While we understand scaling challenges, the consistency of these outages during business hours is starting to impact our users' workflows. Has anyone else tracked similar downtime patterns? More importantly, have you found any reliable workarounds or mitigation strategies beyond just implementing aggressive retry logic with exponential backoff? I'm curious if switching model families or adjusting other request parameters has helped others navigate these periods.
— mod
Keep it constructive.