Nothing provider-wide right now
The test we apply
The trick is that a provider-wide slowdown and a local one leave different fingerprints, and we measure both. A provider-wide slowdown shows up for everyone at once — many independent contributors, or our own probes, all watching the same model see the queue stretch together. A local problem shows up for one account. So the answer is not a guess; it is a comparison you can reproduce:
- 1 · Compare the model to its own recent past.
We take the last hour of completed batch jobs on a model and divide its median wait by the same model’s 30-day median. A ratio near 1 means a normal queue; a high ratio means the queue is genuinely slow right now — this is exactly what GET /v1/conditions returns.
- 2 · Check whether it is slow for everyone, or just you.
If that slowdown is seen across many independent contributors at once, it is the provider — and it becomes a recorded outage. If only your jobs are slow while the model reads normal here, the problem is on your side.
- 3 · Read the confidence before you act.
A brand-new model has too little history to judge yet, and we say so rather than guess. When we have the measurements, we state the verdict plainly; when we don’t, we tell you it is thin — never a fabricated reading.
How we grade the queue
| Verdict | Queue vs normal | What it means |
|---|---|---|
| Much slower than usual | 3x+ | The queue is running at least three times its normal wait. When this shows across many contributors at once, it is them, not you. |
| Slower than usual | 1.5x+ | The queue is running noticeably slower than this model’s own 30-day median — enough to notice, not yet dramatic. |
| Normal | about 1x | The queue is behaving the way it usually does for this model. If your job is slow while this reads normal, the problem is almost certainly on your side. |
A ratio is the recent median queue time divided by the model’s own 30-day median. These are the same thresholds GET /v1/conditions uses — one source, so the page and the API can never say different things.
What we can see right now
No batch queue we watch is in an outage this minute. We are watching 7 provider/model batch queues right now. When one crosses the bar, it appears here and in the outage archive with the measured slowdown, how long it has been open, and who saw it.
Why you can trust the answer
We are watching 7 provider/model batch queues right now. We are the only people in this niche who have measured what everyone else guesses about — real completed batch jobs, from independent contributors and our own probes, not a provider’s marketing SLA. When we say a queue is slow for everyone, we can show the measured slowdown, the baseline it beat, and how many independent sources saw it. When we say the problem is on your side, it is because the queue we watch is behaving normally. Either way, you get an answer you can act on — and cite.
See our own live status · Browse every outage we measured · Read the /v1/conditions reference