Anthropic

Which Anthropic datasets hold data

True for each dataset that has something to show, so the board hides the cards that do not. costs is billed spend, sku and resource are billed spend with a token type or a named key or person, usage_value is Claude Code and Cowork usage rolled into the cost tables, and usage is that usage as it was reported, which is what the usage routes read. people and cowork say more about the reported usage: whether two or more people reported any, and whether Cowork did. list_price is Claude Enterprise spend that carries a list price. activity has one flag per kind of activity Claude Code reported, and fleet, managed_settings, organizations and teams say what its installs reported about themselves. The activity flags and those four look at no range: each is true while anything of it is stored, whatever its day.

Claude Code and Cowork usage, in total and per day

Usage is what Claude Code and Cowork reported about themselves, valued at API list price. It is not what the organization is billed, and it includes the current day. totals covers the range, with each person counted once. daily has one entry per day of the range, at zero on a day nobody used Claude.

Claude Code and Cowork usage by person, by model or by product

Usage is what Claude Code and Cowork reported about themselves, valued at API list price. It is not what the organization is billed, and it includes the current day. Rows are ordered by usage value, largest first. share is a row's part of the whole range, whatever search narrows the rows to. Grouped by person the answer names people, so it is sent with Cache-Control: no-store.

How Claude Code and Cowork usage is spread across people

Usage is what Claude Code and Cowork reported about themselves, valued at API list price. It is not what the organization is billed, and it includes the current day. Each person is counted in the band their usage value over the range falls in, and every band is answered, including one with nobody in it. top_decile and rest set the tenth of the people who used the most beside everyone else.

Claude Enterprise at list price, against what it cost after discount

What each Claude product would have cost at Anthropic's list price, beside what it cost after discount, over the range. The amount after discount is the same whether Anthropic or a marketplace invoices it. Anthropic reports a list price for Claude Enterprise alone, so an organization on another plan is answered with zeros.

What Claude Code was used for, by kind of activity

Activity is what Claude Code and Cowork reported about themselves, kept as daily totals: names, counts and durations, never the text of a prompt, a response or a command. sections has one entry per kind of activity over the range, with its largest rows. response_times estimates, per model, how long a request took and how long its first token took: the times are kept in buckets, so each figure can be off by the width of one. people says how many people each kind of activity reached, and names nobody.

Claude Code activity per day

Activity is what Claude Code and Cowork reported about themselves, kept as daily totals: names, counts and durations, never the text of a prompt, a response or a command. A series per key of the measures the board charts (session, active_time, prompt, permission_mode, api_error, lines, git_output). Every day of the range has a value, zero on a day nothing was reported.

The Claude Code installs of the people who used Claude

The people counted are the ones who reported usage in the range. A profile is kept per person, not per install: someone with two installs, such as a terminal and an editor extension on different versions, is shown as the one that reported last. Versions, terminals, operating systems, managed settings, Claude organizations and team tags, each as a count of people. Someone whose install did not report a value is left out of that list. Names nobody.

The people behind the largest activity figures

Activity is what Claude Code and Cowork reported about themselves, kept as daily totals: names, counts and durations, never the text of a prompt, a response or a command. Each list names at most three people, the largest figure first, and nobody whose figure is zero. behind_on_version and without_managed_settings count everyone they apply to among the people who reported usage in the range, and name three. The answer names people, so it is sent with Cache-Control: no-store.

People for whom Claude has reported nothing lately

Measured from today in UTC, with no range: how many people had nothing reported for them in the last 14 days and in the last 30, and the ten away the longest. Someone is last seen on the latest day any of their usage or activity was reported, and someone last seen more than 400 days ago is not counted. The answer names people, so it is sent with Cache-Control: no-store.

People who started using Claude, came back to it or kept using it

Everyone who reported usage in the range is one of three: new when they were first seen inside it, continuing when they also reported usage in the range of the same length just before, and returning otherwise. At most five are named: up to three of the new and up to two of the returning, each kind taking the places the other leaves empty. The answer names people, so it is sent with Cache-Control: no-store.