AI NewsProductivityReported
A test found Anthropic's Files API cost 125 more tokens than pasting
Anthropic's Files API left beta on 19 August. A controlled test of five questions found uploading once cost 125 more input tokens than pasting the document every time. Prompt caching cut billed input to roughly a third.

Image: The New Stack
Why it mattersUpload once and reference by ID sounds like it should be cheaper. It is not, and the number of teams who will assume otherwise is the reason this test is worth reading.
Anthropic moved the Files API and computer-use toolset out of beta on 19 August. Jessica Wachtel ran a controlled test of the Files API against the obvious alternative and published the numbers in The New Stack on 27 August.
The result is not what the design of the feature suggests.
How it was tested
She wrote a fake API reference for an invoicing company, about 1,200 words covering authentication, rate limits, idempotency, webhooks, bulk endpoints and a sandbox. Then five developer questions the document can answer, each with a specific correct answer including details that are easy to miss, such as token scopes not being editable after creation and the bulk endpoint having its own rate limit.
The same five questions ran three ways against claude-sonnet-5 with identical instructions:
- Paste the full reference into every request
- Upload it once through the Files API, reference the file ID
- Paste it once into the system prompt with prompt caching enabled
All fifteen answers were correct in all three setups. Answer quality did not change.
The numbers
| Setup | Input tokens | Cache write | Cache read | Output |
|---|---|---|---|---|
| 1, paste every time | 15,246 | 0 | 0 | 1,399 |
| 2, Files API | 15,371 | 0 | 0 | 1,434 |
| 3, prompt caching | 271 | 2,990 | 11,960 | 1,642 |
Setup 2 billed more than Setup 1. The uploaded file's contents are still processed into every request, at roughly 3,050 input tokens per question either way, and referencing the file adds about 25 tokens of overhead on top. Across five requests that is 125 extra input tokens, and there is no number of requests at which this reverses.
Setup 3 behaved the way people expect the Files API to behave. The document was billed once as a 2,990 token cache write, then read from cache four times at roughly a tenth of the normal input rate. Cache writes carry a 25 percent premium, so the first request is the most expensive and the saving comes on every request after it.
What each feature is for
The article's conclusion is clear and worth adopting as a rule.
The Files API manages documents. One uploaded copy referenced by ID instead of document text stored in your source code. Formats you cannot paste, like PDFs and images. Expiration settings on stored files. It does not change cost, and Anthropic's announcement never said it would.
Prompt caching lowers cost. In this workload it cut billed input to roughly a third across five requests, and the gap widens with every request after the first.
Two caveats on the caching result
The cache expires after five minutes of inactivity, so the saving assumes requests keep arriving. A workload with long gaps pays the write premium repeatedly and may end up worse off than pasting.
Caching also required restructuring the request, moving the document into the system prompt with a cache marker. That is a code change, not a configuration flag.
One incidental finding worth knowing: temperature is deprecated on claude-sonnet-5 and the API rejected temperature=0.
Source
- Anthropic's new Files API vs. pasting: It will save you time, but it won't save you money., The New Stack, 27 August 2026
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

