Anthropic launches Prompt Caching Dashboard, visualizing cache hits and costs

robot
Abstract generation in progress

ME News, on April 22 (UTC+8), according to Dongcha Beating monitoring, the Claude Developer Console has launched the Prompt Caching Dashboard, accessible at platform.claude.com/usage/cache. The dashboard can be filtered by workspace, model, and time period, and mainly shows three sets of data: cache read ratio, which is the proportion of requests that hit existing cached content; cache usage composition, which breaks input tokens into four categories—uncached, 5-minute cache write, 1-hour cache write, and cache read—and displays them as stacked bar charts; and write amortization, which measures how many times a single cache write is reused in subsequent reads. In the screenshot example, Claude Opus 4.6 processed 27.4 hundred-million (2.74 billion) input tokens within 7 days, with a read ratio of 85.4% and a write amortization of 8.65x. At the bottom, there is also a time-series chart of cache read ratio with granularity ranging from 1 hour to 24 hours. Anthropic’s prompt caching mechanism allows API users to mark cacheable content such as system prompts and long contexts. The first cache write is charged an additional fee, while subsequent cache hits are billed at about one-tenth of the standard input price. By default, cache retention is 5 minutes, and paid options can extend it to 1 hour. Previously, users could only indirectly judge cache effectiveness by the token count field returned by the API, without any visualization tools. (Source: BlockBeats)

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin