Is the OpenAI batch queue predictable?

We measure it: gpt-5.6-luna runs a spread of 7.6x (p90 10 min over p50 82s), so the tail sits a known distance from the median. A tight spread means the median is safe to book a deadline against - which is a real part of whether the batch API is worth it, not just the 50% price cut.

Is the OpenAI batch API worth the wait?

For most workloads, yes. It is 50% cheaper, and half of gpt-5.6-sol jobs finished within 30s in our measurements - well inside the 24 hours window OpenAI promises. The only question is your deadline, which is why batchwatch answers against a max_wait you supply rather than guessing one.

How much does the OpenAI batch API save?

50% of the synchronous price, on both input and output tokens. That is OpenAI's own published batch rate, not a batchwatch estimate.

Is the OpenAI Batch API worth it?

The answer

OpenAI's batch endpoint costs 50% of the synchronous rate - their own published price, on every input and output token. That is a real, immediate saving on any workload that does not need an answer this second.

And the wait is shorter than the "within 24 hours" ceiling suggests: we measured a median of 30s for gpt-5.6-sol, with nine in ten jobs finishing within 16 min. For most pipelines that is an easy trade - half the cost for a wait you can plan around.

There is a second half to "worth it" that price alone misses: how predictable the queue is. gpt-5.6-luna's spread - its p90 over its p50 - is 7.6x: the tighter that number, the more a median is safe to plan on. batchwatch measures it per model, so you can pick on consistency, not just on the headline wait.

Every model we have measured

Every OpenAI model we can stand behind a percentile for - price is the same 50% cut on all of them; the wait is what differs. Most predictable first - spread is p90 over p50.

Model	Median (p50)	p90	Spread	n
OpenAI gpt-5.6-luna	82s	10 min	7.6x	3390
OpenAI gpt-5-nano	74s	25 min	20.3x	3899
OpenAI gpt-5.6-sol	30s	16 min	32.4x	1884

Each row is measured over the last 30 days, and links to that model's own page with its full distribution and confidence grade. A model we have not yet measured enough of to stand behind a percentile is left out entirely - never filled in with an estimate.

The full numbers, per model

Questions this answers

Is the OpenAI batch queue predictable?: We measure it: gpt-5.6-luna runs a spread of 7.6x (p90 10 min over p50 82s), so the tail sits a known distance from the median. A tight spread means the median is safe to book a deadline against - which is a real part of whether the batch API is worth it, not just the 50% price cut.
Is the OpenAI batch API worth the wait?: For most workloads, yes. It is 50% cheaper, and half of gpt-5.6-sol jobs finished within 30s in our measurements - well inside the 24 hours window OpenAI promises. The only question is your deadline, which is why batchwatch answers against a max_wait you supply rather than guessing one.
How much does the OpenAI batch API save?: 50% of the synchronous price, on both input and output tokens. That is OpenAI's own published batch rate, not a batchwatch estimate.

See it live

Every number here is measured, and it is all public.

The live dashboard shows every model with its distribution, n and confidence grade. /v1/coverage is the same data as JSON, no key required. Send five measurements in seven days and the delay is gone: you read every figure live, and the decision endpoints open (/v1/should-i-batch, /v1/estimate-batchtime).