The answer
We have measured this. Across the 1 Anthropic model we have enough completions for, the median wait ranges from 2 min (claude-haiku-4-5) to 2 min (claude-haiku-4-5). Nine in ten claude-haiku-4-5 jobs finished within 5 min. Every figure carries n and a confidence grade on its own page.
Anthropic promises only "within 24 hours" - a ceiling, not a typical wait. The measured medians above are what jobs actually get, and they are why batching is worth the wait for most workloads.
How predictable is it?
2.4x spread - p90 over p50
How much longer the slow tail runs than the typical job. Nearer 1× is more predictable.
Half of claude-haiku-4-5 jobs finished within 2 min, and nine in ten within 5 min - a spread of 2.4x. The slow tail runs 2.4x longer than the typical job, so planning on the median alone would set your deadline that far too tight. That gap is exactly what a median hides - and it is the number no provider publishes. The tightest we measure is gemini-3.7-flash at Google (1.3x); this one sits at 2.4x.
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
- How long does the Anthropic batch API actually take?
- Half of claude-haiku-4-5 jobs finished within 2 min and nine in ten within 5 min, across 2608 measured completions in the last 30 days.
- Is that faster than Anthropic's 24-hour window?
- Far faster for the typical job. Anthropic guarantees completion within 24 hours; the measured medians above are minutes, not hours. The 24-hour window is the contract, not the experience.
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).