# Article Name How to Monitor Claude Billing and Anthropic API Spending (3 Methods) # Article Summary Explore three practical methods to monitor Claude and Anthropic API spending, from the Claude Console to the Admin API, and avoid unexpected charges # Original HTML URL on Toriihq.com https://www.toriihq.com/articles/how-to-monitor-spending-claude # Details Keeping tabs on Claude billing in your Anthropic account can save you from surprise charges and burned credits. It's easy to lose track when multiple workspaces, API keys, and models run at once, and agentic workloads like Claude Code can consume tokens far faster than a chat interface. Model choice alone swings the rate by 5x: Anthropic lists Claude Opus 5 at $5 per million input tokens and $25 per million output tokens, against $1 and $5 for Claude Haiku 4.5 (source: https://platform.claude.com/docs/en/about-claude/pricing). This article walks through three practical ways to read the Claude Console's Usage and Cost pages, set spend limits, and query the Anthropic Admin API's usage and cost endpoints directly so you don't get hit with an unexpected charge. ## Use the Claude Console Usage and Cost Pages Use the Claude Console (platform.claude.com, formerly console.anthropic.com) to view, filter, and export usage and cost data so you can track Claude billing without touching the API. ### Open the Usage page - Sign in to the Claude Console and open the Usage page from the left sidebar. - The page charts token consumption over time, which is the quickest place to spot a spike from a runaway script or an agent loop. ### Open the Cost page for dollar amounts - The Usage page shows tokens; the Cost page shows dollars. Open Cost to see spend charted over your selected date range. - Use the group-by controls to break spend down by workspace, API key, or model. Scan for expensive models or specific days with big jumps. - If your organization uses workspaces (for example one per team or environment), the workspace breakdown is the fastest way to attribute a spike to an owner. ### Pick the date range and filters - Use the date selector to choose recent days, the current month, or a custom range. - If you want monthly reports, set the custom range to the billing period. - Look at the top-line totals over your selected range to get a quick sense of where you stand. ### Export detailed usage for analysis - From the Usage or Cost page, use the export option to download a CSV of the data behind the chart. - Load the CSV into a spreadsheet or BI tool for charts and internal reporting. ### Set spend limits so overruns stop themselves - In the Console settings you can set organization spend limits, and each workspace can carry its own monthly spend limit. Anthropic's workspace documentation is explicit about the hierarchy: "You can set workspace limits lower than (but not higher than) your organization's limits" (source: https://platform.claude.com/docs/en/manage-claude/workspaces). - Workspace limits are cheap insurance: a capped dev workspace cannot quietly consume the production budget. ### Review invoices and payment settings - Open Settings, then Billing, to find invoices, your current credit balance, and payment methods. - Download invoices for accounting and match them to your exported cost data if you need line-item detail. - If you prepay with credits, check the remaining balance here so a depleted balance never interrupts production traffic. ### Communicate findings and set a routine - Save the CSV and invoice PDFs to your finance folder and share the key numbers with stakeholders. - Run the Cost page monthly (or weekly if spending is high) and export a short report so surprises are rare. ## Use Torii for Automated Claude Spend Monitoring Rather than checking the Claude Console manually each month, you can use Torii, a SaaS Management Platform, to centralize Claude cost monitoring alongside every other SaaS and AI tool. Torii's Claude spend management feature surfaces your Claude usage and spend next to the rest of your AI portfolio, so finance, IT, and security teams share a single view. 1. Sign up for Torii: contact Torii and ask for your free two-week proof-of-concept. 2. Connect your Anthropic account to Torii using the Claude Developer integration, which connects to your organization, workspaces, and members. 3. Monitor Claude spend in the Torii dashboard: you'll see total spend, token consumption, and active users at a glance, with breakdowns by model, team, and top users, so you can spot a spike and its owner in the same view. ## Query the Anthropic Admin API Usage and Cost Endpoints Call Anthropic's Admin API to pull Claude API billing data programmatically. Anthropic's documentation is blunt about the credential: "These endpoints require an Admin API key, which is different from a standard Claude API key" (source: https://platform.claude.com/docs/en/manage-claude/usage-cost-api). The key starts with sk-ant-admin, and an organization admin creates it in the Console under Settings. A regular API key will not work. ### Choose the endpoints to call - GET /v1/organizations/usage_report/messages returns token usage (uncached input, cache reads, cache writes, output) bucketed over time, with optional grouping by API key, workspace, model, or service tier. - GET /v1/organizations/cost_report returns cost amounts in USD, bucketed by day, with optional grouping by workspace and line-item description. Use the two together: the usage report tells you which keys, models, and workspaces are consuming tokens, and the cost report converts activity into dollars. The split between those token categories matters, because Anthropic prices a cache read at 0.1x the base input rate and a 5-minute cache write at 1.25x (source: https://platform.claude.com/docs/en/about-claude/pricing), so two workloads with identical token counts can land a long way apart on the invoice. ### Make authenticated requests Set your Admin API key in an environment variable. The Admin API uses the x-api-key header plus an anthropic-version header (for example anthropic-version: 2023-06-01). Query api.anthropic.com with starting_at and ending_at timestamps in RFC 3339 format, plus optional bucket_width and group_by[] parameters. ### Parse the responses and sum the amounts Both endpoints return a data array of time buckets, each with a results list. The cost report returns each amount as a decimal string in USD, so parse it as a decimal rather than a float before summing. ### Attribute spend with group_by - Add group_by[]=api_key_id or group_by[]=workspace_id to the usage report to see which key or team is driving consumption, and group_by[]=model to catch an expensive model doing work a cheaper one could. - Run the same review against Anthropic's batch pricing. Its pricing page states that "The Batch API allows asynchronous processing of large volumes of requests with a 50% discount on both input and output tokens" (source: https://platform.claude.com/docs/en/about-claude/pricing), so a non-urgent workload still running on the standard endpoint is paying roughly double. - These endpoints group by key, workspace, and model, not by person. If several developers share one key, the report cannot tell you who spent what, which is a common reason teams move each developer or service to its own key or workspace. ### Handle pagination and rate limits - If a response paginates, follow its has_more flag and pass the returned next_page value on the next request until done. - Respect 429 responses: back off, then retry with exponential backoff. ### Automate daily checks and simple alerts - Run the cost report for yesterday each morning to catch unexpected daily spikes. - Calculate rolling totals (7-, 30-day) by querying with different date ranges and compare them to your budget thresholds. - If totals exceed thresholds, trigger an alert from your system. - If your team uses Claude Code, Anthropic also exposes a dedicated Claude Code Analytics API with per-user daily metrics. ### Verify results against the Console periodically Use the same date ranges and compare your API totals to the Console's Cost page to confirm your parsing is correct. If numbers don't match, re-check the date boundaries (the API uses UTC timestamps) and whether credits or grouped line items explain the gap. That's the flow: call the usage report for token consumption, pull the cost report for dollars, group by key or workspace to attribute spend, handle pagination and rate limits, and run these checks on a schedule to catch surprises.