Batch processing

Half price, if you can wait.

The rates

Model	Batch input	Batch output
Claude Fable 5	$5	$25
Claude Opus 5	$2.50	$12.50
Claude Sonnet 5	$1	$5
Claude Haiku 4.5	$0.50	$2.50

Per million tokens. Exactly 50% of the synchronous rate, on both directions.

The other reason to use it: 300K output

On the Batch API, Claude Opus 5, Opus 4.8, Opus 4.7, Opus 4.6, Sonnet 5 and Sonnet 4.6 support up to 300K output tokens with the output-300k-2026-03-24 beta header — against 128K synchronously.

If you need a single very long generation, batching is not only cheaper, it is the only route to that ceiling.

What it does not work with

• Fast mode. Opposite trades; not combinable.
• Managed Agents. Sessions are stateful and interactive; there is no batch mode.
• Interactive latency. Obviously.

It does stack with prompt caching.

Good candidates

Bulk classification, evaluation runs, dataset labelling, back-catalogue summarisation, overnight report generation — anything where a delay costs nothing and volume costs a lot.