Is OpenAI down right now?

While you wait: the OpenAI incident checklist

1. Split ChatGPT from the API. ChatGPT and the API are separate components with separate incidents. Your product cares about api.openai.com — check which component, and which models, the incident actually names.
2. Check your error mix. 429s vs 5xx vs timeouts tell different stories. Tune retries to exponential backoff with jitter and a hard cap, and keep operations idempotent — a retry storm amplifies their outage into yours.
3. Flip to degraded mode deliberately. Fallback model, alternate provider, cached responses, or a clean 'AI features temporarily unavailable' state. Any of these beats spinners and timeouts bleeding into the rest of the product.
4. Protect your own capacity. LLM calls that hang until timeout tie up workers and connections. Shorten timeouts during the incident and shed AI load early so the non-AI product stays healthy.
5. Pause the async backlog. Batch and queue consumers hammering a degraded API just build a failure backlog. Pause or slow them; drain once the status page clears.

What breaks when OpenAI has issues

An OpenAI incident lands wherever you call the API: chat features, summarization, embeddings and search pipelines, agents, and background enrichment jobs fail or crawl. The second-order damage is usually self-inflicted — generous timeouts and eager retries exhausting your own worker pools, and queues backing up faster than they drain. ChatGPT being down is an annoyance for your team; the product impact comes from the API side.