How to see Claude Code, Codex and Cursor spend in one place, and set a budget across a team

Claude Code on a Team or Enterprise plan

Anthropic's Claude Code analytics are at claude.ai/analytics/claude-code, and "Admins and Owners can view the dashboard". It shows "lines of code accepted, suggestion accept rate, daily active users and sessions". Once a GitHub admin installs Anthropic's GitHub app, it also shows pull requests made with Claude Code. A leaderboard shows "the top 10 users ranked by contribution volume", and you can export every user's figures as a CSV file. The programmatic Analytics API "is not available on the Teams plan".

Money is on a different page. Anthropic's spend report "shows estimated spend per user and per model, with CSV export, updated daily", but it covers only usage credits, which are what a member uses after the seat's allowance runs out: "Usage inside the seat allowance isn't metered in dollars." The data "refreshes daily and has a one-day delay", and on seat plans "the export only reflects spend that exceeds your seat allotment." So an admin can see who is active and what each person spent beyond their seat, but not a dollar figure for the seats themselves; that figure is the seat price times the number of seats.

Claude Code through the API

Teams that pay per token get a separate dashboard in the Claude Console, which shows "Spend: daily API costs in dollars" and a table of spend per person. Anthropic adds that "Spend figures in the Console dashboard are estimates for analytics purposes." Admins can set workspace spend limits on total Claude Code spend in the Console. This is the clearest dollar view Anthropic offers, and it exists only for teams that pay per token rather than per seat.

Codex on ChatGPT Business, Enterprise or Edu

OpenAI's governance page draws the line: "ChatGPT provides workspace-wide analytics for broad adoption and engagement. Codex analytics focuses on Codex activity." For reporting that lasts, it recommends the Analytics API. Its usage insights, under Analytics > Insights in the Admin Console, show messages, "Share of credits" and active users by category of task, and OpenAI warns that "A message count doesn't tell you how much work the team finished."

The unit is credits, not dollars. OpenAI's pricing page gives each model's rate in credits, for example 50 credits per million input tokens and 250 credits per million output tokens for GPT-6.1 Sol. OpenAI's usage-limits page says "ChatGPT workspace usage limits and spend controls apply to eligible activity under your workspace plan, which can include some Codex activity", and sends admins to OpenAI's help centre for the details.

The help centre, read in a browser on 4 November 2026, fills in the Business detail. Standard and Premium seats include usage for Codex, and "When that included usage is exhausted, workspace credits can cover additional eligible usage, subject to your workspace's spend controls". Owners and admins can set monthly credit limits per seat type and a per-user override that takes precedence, and "By default, all seats and users have no limits specified" (Managing credits and spend controls in ChatGPT Business). Workspaces that had Codex seats before 24 June 2026 can keep those legacy seats, which draw on credits from their first use.

For reporting, the Admin Console's Analytics section "can include ChatGPT or Codex usage, active users, leaderboards, credits, and exportable activity", depending on role and plan. Codex usage and credit analytics go back up to 120 days and refresh about every 1 to 6 hours, and "Estimated costs, when shown, are planning estimates rather than settled charges or invoices" (Managing your tenant in Admin Console).

Cursor on a Teams or Enterprise plan

Cursor's team analytics are "available for Team and Enterprise customers" and show each person's chats, accepted Tab suggestions and accepted agent lines, for up to 10 people and 90 days at a time. That page says nothing about money. Spend is on the billing side: admins set "one monthly additional usage limit for the whole team" plus a default limit per member, and spend alerts send an email when the team's additional usage passes a threshold, though "Spend alerts don't stop usage". Both cover only the usage beyond each seat's weekly pool, which is the same gap as Anthropic's spend report.

What none of them shows

Each dashboard stops at its own tool. None of them shows the other two, and none adds the seats and the extra usage into one monthly figure per person across vendors. A lead whose team pays for two or three of these tools ends up with three logins, three units (dollars of usage credits, OpenAI credits and dollars of Cursor additional usage) and no total.

OpenReason Intel puts Claude Code and Codex in one view. Each engineer runs one command from the or-otel quickstart on their Mac, Windows or Linux machine. It sets up a small reporter that sends the usage each tool already records to the team's OpenReason account. The requests themselves still go straight to Anthropic and OpenAI on the team's own plans, so OpenReason never sits in between.

• The Spend page shows seat commitments and variable spending together, with vendor totals, Claude usage credits and Codex credits. It labels which figures are committed prices, which are amounts the vendor reported, and which are estimates.
• The People page shows each person's usage, their 5-hour and weekly window history, the periods when a window ran out, and their spending, side by side with the rest of the team.
• The Work page shows branch and feature activity beside agent time, turns, window use, usage beyond the plan, and recorded pull request links, so you can inspect a figure against the work recorded with it.

The figures come from usage data on each engineer's machine, so they are a view of spending rather than a reconciled invoice. It is free for teams under 10 engineers and $20 an engineer a month above that, with no card to start (pricing). It reads Claude Code and Codex only; OpenCode support is coming.

A budget across tools is three decisions, and each uses something that exists today.

1. Pick the number. Anthropic publishes a starting point for Claude Code: "Across enterprise deployments, the average cost is around $13 per developer per active day and $150-250 per developer per month, with costs remaining below $30 per active day for 90% of users." That figure is for Claude Code alone, paid through the API, so a team on seats starts from the seat price and decides how much usage beyond the seat it will accept on top. Neither OpenAI nor Cursor publishes a comparable average that we could find.
2. Set each vendor's own cap on usage beyond the seat. Anthropic lets Owners set spend limits on usage credits "at the organization, group, or individual member level". On ChatGPT Business, owners and admins set monthly credit limits per seat type, with per-person overrides (OpenAI's help centre); no limit is set by default. Cursor lets team admins set a team-wide monthly limit on additional usage and a default limit per member. Each cap covers its own tool and knows nothing about the other two.
3. Compare the running total with the number in one place. For Claude Code and Codex, OpenReason Intel's Spend and People pages add up the seats and the variable spending per person, so you can see who is near or past the number without opening two admin consoles. It shows the figure; it does not stop spending or send a warning. Because the figures are in the dashboard, a team with a coding agent can build its own warning on top.