The answer
The trade is simple and it is yours to make. Batch costs 50% of the synchronous rate at OpenAI, Anthropic, Google - their published price - in exchange for a queue wait. Synchronous costs double and answers immediately. The only variable is your deadline.
What tips it toward batch for most workloads is that the wait is short: we measured a median of 51s for gpt-5.6-sol, with nine in ten jobs finishing within 5 min. Against half the price, minutes of wait is an easy call unless you are on a hard, immediate deadline.
Every model we have measured
Every model we have measured enough of to publish a percentile, across every provider. Most predictable first - spread is p90 over p50.
| Model | Median (p50) | p90 | Spread | n |
|---|---|---|---|---|
| Google gemini-3.7-flash | 2 min | 3 min | 1.3x | 2298 |
| Anthropic claude-haiku-4-5 | 2 min | 4 min | 2.2x | 2583 |
| OpenAI gpt-5.6-sol | 51s | 5 min | 5.8x | 1646 |
| OpenAI gpt-5.6-luna | 79s | 9 min | 6.6x | 3230 |
| OpenAI gpt-5-nano | 73s | 30 min | 24.3x | 3863 |
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.
How predictable is it?
1.3x spread - p90 over p50
How much longer the slow tail runs than the typical job. Nearer 1× is more predictable.
Half of gemini-3.7-flash jobs finished within 2 min, and nine in ten within 3 min - a spread of 1.3x. The tail sits close to the middle, so what you measure once is close to what you get again, and sizing a deadline against the tail costs almost nothing. It is the tightest distribution we measure across every model on the site - the most predictable batch queue we have found.
Spread is p90 divided by p50, both measured here over the last 30 days. No provider publishes it; it is the number that tells you whether a median is safe to plan on.
The full numbers, per model
Questions this answers
- Should I use the batch API or synchronous?
- Use batch whenever the work can wait: it is 50% cheaper at every major provider. Use synchronous only when you need the answer immediately. batchwatch measures the real batch wait so you can size that "can wait" against actual numbers rather than the 24-hour ceiling.
- How much does the batch API save over synchronous?
- 50% - batch is half the synchronous price on both input and output tokens at OpenAI, Anthropic and Google. That is their published price, 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).