Everything we measure is free. Contributing makes it live.

Free, for everyone, no account

Every queue time batchwatch has measured is open to read. The whole dashboard, every model page, the comparison tables, the outage feed and the CSV export - no account, no email, no card. The read API answers the same measurements 15 minutes behind live, and a new key comes with 20 live calls to try it with.

These are measurements, not estimates. Every percentile on this site came from a job that was actually submitted, actually queued and actually finished on the provider's own batch API. That is the part nobody else has, and it is the part we give away.

Contribute, and the delay disappears

Send your own batch timings and you read the queue as it is, not as it was 15 minutes ago. That is the trade, and it is the right one: the people whose measurements build the dataset should not be reading a delayed copy of it. A queue that looked fine 15 minutes ago is a dashboard. A queue that looks fine now is something you can route production traffic on.

You cannot set your own tier - it is measured in the data you send. That is not a hurdle: a dataset where accounts can promote themselves is worth nothing to read, and being worth reading is the entire product.

How to contribute, exactly

1. Get a key. POST /v1/keys - one command, no email, no password, no confirmation step. Full recipe at /docs/keys.
2. Send finished jobs. POST /v1/calls/complete takes a batch job you have already run - model, token counts, submitted-at and finished-at - in a single request. Jobs still running can be timed live instead with POST /v1/calls and PATCH /v1/calls/{id}. Both are in /docs/ingest.
3. Keep it current. 5 measurements inside 7 days holds live access - roughly twice a week, not once a day. We are asking you to stay current, not to change how you work.

If you already run batch jobs the timings are sitting in your logs, and the client libraries send them in two lines - nine languages at /clients. Measure on separate days and your figures start counting towards the published percentiles, so no single account can buy influence with volume alone.

Verify an email - double the volume, free

10,000 calls a week instead of 5,000, for confirming an address. Contribution already made your figures live, so the email buys headroom - and it is how we reach you when a queue you depend on breaks or your key is about to stop working. It is the only personal data we ask for anywhere, and it stays separable from your measurements.

Verify an email

Is batching worth it

Batch pricing is about 50% off synchronous at every major provider. If roughly 30% of your inference can wait, that is 30% of the bill cut in half - and the only question left is how long the queue is, which is the question this site answers with measurements instead of guesses.

That 30% is the one number you bring: your workload's batchable share, yours to set, with a deliberately conservative default - most teams can batch more. Everything else here is measured. Put your own share into the worked example and the arithmetic is yours.

Your measurements stay yours

Everything you send comes back out again: GET /v1/calls/mine exports it and DELETE /v1/calls/mine removes it, both from the key itself. The published aggregate is licensed CC BY - reuse it, including commercially, for the price of a credit and a link back. The details are in the terms and the privacy note.

Nothing here is for sale

There is no plan, no card and no invoice. A measured dataset gets better the more people feed it, so the only currency that buys anything here is the one the dataset actually runs on: measurements. Contribute and you get the fastest version of the answer. Read for free and you get the same answer, slightly later. Both are a good deal, and neither needs a purchase order.

Holding a subscription from before this? It keeps working exactly as it did, and you can manage it any time through POST /v1/billing/portal. Anything else - volume beyond these numbers, a private deployment, or a question - is hello@batchwatch.dev.