How much does AI document summarisation cost?
Turning long uploaded documents into a short summary or structured extract.
The short answer
On Claude Haiku 4.5, at 40,000 input and 800 output tokens per call. Forty thousand input tokens is roughly a thirty-page document. The output is deliberately small: the whole point of the feature is that the summary is shorter than the source.
Your numbers
Now put your own in
The fields open with the profile above. Change them to match your feature and the read updates as you type.
Your numbers
Rough numbers are fine. You are checking whether you are near the line, not being exact.
Classification, extraction, routing. Anything where the job is narrow and the volume is high.
Everything you send it
What it sends back
How often one user uses it
What it costs you
What drives the cost
Input, almost entirely, and it scales with document length rather than with anything you control. A user who uploads a 400-page contract costs ten times what your model assumed.
Where the estimate goes wrong
Pricing on the average document. Document length has a long tail, and the tail is where the cost is. Cap the input length explicitly, chunk beyond the cap, and decide what the product does when somebody uploads something enormous, because they will.
What this does not price
Summarisation is the use case where the cheap model is most often good enough, and where teams reach for the frontier model out of habit. Before accepting a verdict of do-not-build here, run the same document through the cheapest model in the list and read both outputs side by side.
Other features people cost
Next step
If document summarisation is on your roadmap and the number here made you pause, that is exactly the conversation worth having before an engineer starts.