CodeCost vs Claude Code Analytics: what Anthropic’s dashboard shows, and what it can’t

If your organisation is on Claude for Teams or Enterprise, you already have a Claude Code analytics dashboard. It is well made and it is free. This is an honest account of what it answers, what it structurally cannot, and why CodeCost is built to be the one dashboard for your AI coding spend.

By the CodeCost team · 8 min read

What Claude Code Analytics actually is

There are two dashboards under the name, and it helps to keep them apart. Everything below is from Anthropic’s own documentation.

For Claude for Teams and Enterprise

At claude.ai/analytics/claude-code, admins and owners get:

• Usage metrics — lines of code accepted, suggestion accept rate, daily active users and sessions.
• Contribution metrics — merged PRs and lines of code shipped with Claude Code assistance, once you install the Claude GitHub app. Anthropic is careful to call these a conservative underestimate.
• A leaderboard of the top contributors by Claude Code usage, exportable as CSV.
• Spend, indirectly — per-user token counts and estimated usage-credit spend come from a spend report you export from the organisation’s analytics settings, or from configuring OpenTelemetry export on each machine. Spend is not on the dashboard itself.

Since July, Enterprise (not Teams) admins also get spend limits at the organisation and user level, alerts at 75% and 90% of the org limit, and model entitlements per role. The Enterprise Analytics API returns per-user usage and cost reports; it is not available on Teams.

For API customers

At platform.claude.com/claude-code, Console users with the UsageView permission see accepted lines, accept rate, a daily active users chart, daily API spend in dollars, and a team insights table with each member’s spend and accepted lines for the current month. Anthropic notes that these spend figures are estimates; the billing page is the truth.

That is a genuinely useful product. If your question is “is Claude Code changing how much we ship?”, the contribution metrics are the best answer anyone has, because only Anthropic can match session output against merged PR diffs.

Where it stops

The gaps are not oversights. They follow from what the dashboard is: one vendor’s view of one vendor’s tool, built to show that tool’s value. Five of them matter for anyone who owns a budget.

1. It only sees Claude

Most engineering orgs in 2026 run more than one coding tool. A developer who uses Claude Code for a refactor in the morning and Codex CLI for a review in the afternoon is one person with one total, and no vendor dashboard will ever show it. OpenAI’s Global Admin Console has the mirror-image blind spot. You end up with two leaderboards, two spend exports in two currencies of “credits”, and a spreadsheet.

CodeCost reads the session files of Claude Code and Claude Desktop, Codex CLI and Desktop, VS Code, Gemini CLI, OpenCode and Antigravity on the developer’s machine, converts every one to dollars with the same pricing tables, and shows one total.

2. It does not know what repository the cost came from

The dashboard attributes to users. But cost is explained by codebases. When most of a team’s spend lands in one legacy service, the useful conversation is about that service — its size, its tests, the migration it is undergoing. CodeCost carries the repository name (never the file paths) on every event, so cost by repository is a first-class view: platform/auth-service $3,120 · 78% of team is the kind of readout you get.

3. It reports; it does not forecast

Spend limits and 75/90% alerts are caps. They tell you that you have arrived at a threshold, and on Teams you do not get them at all. What a budget owner wants on the 8th of the month is whether the month is on pace for 90% or 130%. CodeCost projects the month from the current pace from the first days and alerts on the projection — Sep 18 · on pace for 108% · alert sent — while there is still time to do something about it.

4. Managers get nothing scoped to them

Access to the Anthropic dashboard is an admin and owner role for the whole organisation. There is no view a team lead can be given that shows their team and nobody else’s. In CodeCost, scoping is the product: the CTO sees the org, a manager sees their team, a developer sees their own usage and a budget in the menu bar. Each team sees its own share, nobody else’s.

5. It updates daily, not live

Anthropic’s analytics work on a daily rhythm: the dashboards chart daily figures, contribution data typically appears within 24 hours of being enabled and then updates daily, and the Analytics API returns daily per-user reports. That suits an adoption review and is too slow for a budget — a runaway session this morning shows up in tomorrow’s numbers. CodeCost is live. Sessions appear as they happen, the month-to-date spend refreshes every ten seconds, the console keeps a live feed of sessions across your tools, and each developer sees today’s spend in their menu bar as it accrues.

Side by side

The privacy difference

Anthropic already has your prompts; its dashboard adds no new exposure. A third-party tool does, so it has to earn that. CodeCost’s answer is a published whitelist: exactly 36 fields leave the machine — tokens, model, timestamps, latency, cost, session and repository identifiers — and prompts, responses, API keys, file paths and the developer’s email never do. The agent is a signed and notarized macOS app that reads files your tools already write; it is not a proxy, it changes no environment variable, and if it stops, your tools keep working. That is what we mean by fail open.

When to use which

If Claude Code is the only AI coding tool your org runs, Anthropic’s own dashboard covers adoption well. However, if you need to see the spend itself — live, across your tools, by team and repository, against a budget and with a forecast for the month — then CodeCost is the tool you are looking for.

It also keeps up as your tools change. CodeCost reads Claude Code and Claude Desktop, Codex, VS Code, Gemini CLI and Antigravity, and open-source harnesses such as OpenCode — so when your developers run open-source models through them, that usage shows up in the same view as everything else. One dashboard, whichever tools and models your team picks next.

For the broader picture — where the numbers live and the four ways to collect them — read How to track AI coding spend across your engineering org.