MetaBench / Workspace
Claude cache lab
Explore fixed examples, or choose live mode to test your own context and question.
Checking API availability…Configure the experiment
Fixed illustrative example. Prompt and question are read-only; sample results do not depend on the displayed text. Choose live mode to test your own input.
Maximum 22,000 characters. Short-prefix mode uses the first 200. The changed-prefix test automatically adds text before the cache breakpoint.
Need a live access token? Request live access by email ↗. Include your model, prompt pattern and expected run volume. Sample experiments do not need a token.
Do not paste an Anthropic API key here. Do not submit secrets or confidential data.
Run report
READYStart with the sample.
Your results will appear here.
Your results will appear here.
—Replay cache-read tokens
—Completed requests
—Estimated run total
| Request | Input | Write | Read | Elapsed | Input cost |
|---|