How to Track Your Token Usage on Claude

2026-11-02

Key Takeaways

On paid Claude plans, open Settings and then Usage (claude.ai/settings/usage) to see your current five-hour session, weekly limits, any per-model weekly caps, reset times and usage credit balance.
Usage is shared across the Claude apps, Claude Code and the desktop app, so heavy coding sessions count against the same allowance as chats.
In Claude Code, the /usage command shows estimated session cost, tokens by model (input, output, cache reads and cache writes), plan usage breakdowns and prompt cache statistics.
API users can track tokens in every response’s usage object and in the Claude Console’s Usage and Cost pages, or pull the data through the Usage and Cost Admin API.
Team and Enterprise administrators get a Claude Code analytics dashboard, downloadable spend reports and an Analytics API.
Thinking tokens count as output, and recent Claude tokenizers produce about 30% more tokens for the same text than older models, so budgets based on older models may run short.

How to Track Your Token Usage on Claude - SentiSight.ai

Image source: Anthropic

You can track your token usage on Claude in three main places, depending on how you use it. Subscribers to Pro, Max, Team and Enterprise plans check Settings, then Usage, in the Claude app or at claude.ai/settings/usage to see how much of their session and weekly allowance they have used.

Developers in Claude Code type /usage to see token counts and estimated costs for the current session. API customers use the usage data returned with every request and the Usage and Cost pages in the Claude Console.

Each view answers a slightly different question. The app shows how close you are to a limit. Claude Code shows where tokens went during a working session. The Console shows what an organization is spending, by model, key and workspace.

The sections below explain what each one shows and how to use the information to spend fewer tokens.

Where to check Claude token usage

Where you use Claude	Where to check	What you see
Claude app on web, desktop or mobile (paid plans)	Settings, then Usage	Session and weekly usage, per-model caps, reset times, usage credits
Claude Code	/usage command, status line, /context	Session cost estimate, tokens by model and type, cache stats, plan usage attribution
Claude API	Usage object in each response; Claude Console Usage and Cost pages	Tokens per request; totals by model, date, API key and workspace
Team and Enterprise administration	Claude Code analytics dashboard, spend reports, Analytics API	Usage and spend across users and teams
Claude through Amazon Bedrock or Google Vertex AI	The cloud provider’s billing console, OpenTelemetry or an LLM gateway	Spend under your cloud account

How to check usage in the Claude App

On the web and in the desktop app, open Settings and choose Usage. In the mobile app, the same page sits under Settings as well. The page shows your progress through the current five-hour session window, your weekly limit, and separate weekly caps for specific models where they apply, such as Claude Fable.

Each bar shows when it resets. If your plan includes usage credits for going past your limits, the page also shows your credit balance and any monthly spend limit you set.

Third-party guides have also described a pace forecast and a breakdown of usage by product, separating Claude Code from chats and other agent work. Layouts change as Anthropic updates the app, so expect some variation.

Anthropic does not publish fixed message counts for each plan, because the number of messages you can send depends on message length, attached files, the model and how much the model reasons.

Free plan users have less to see. Since August 2026, the Usage page on free accounts has been reported as largely empty, according to third-party trackers. Free users mainly learn about limits from in-app warnings when they approach them.

How to track token usage in Claude Code

The /usage command

Typing /usage in Claude Code opens a detailed breakdown of the current session. It shows an estimated total cost at list prices, API time, wall-clock time, lines of code changed and token usage per model, split into input, output, cache reads and cache writes.

Organizations can set custom pricing through a managed setting so the estimate matches their contract. Running /clear resets the session counters in recent versions.

Newer versions of Claude Code add more detail. A prompt cache line shows how much input was served from cache. A plan usage breakdown attributes consumption to skills, subagents, plugins and MCP servers, and offers recommendations, with a shortcut that applies suggested fixes.

Users can toggle between 24-hour and seven-day views. Subscribers with usage credits see a credits row, and a separate /usage-credits command manages them.

Other built-in tools

The spinner shows a running token count while Claude works. The /context command reveals what is taking up space in the context window, which helps spot oversized instruction files or unused tool definitions.

A configurable status line can display tokens or cost at all times. The /insights command generates an HTML report of usage patterns, saved locally in the Claude data folder. Teams that want central monitoring can export Claude Code metrics through OpenTelemetry.

What Claude Code typically costs

Anthropic’s documentation puts average enterprise spending on Claude Code at about $13 per developer per active day, or $150 to $250 per developer per month, with 90% of users spending under $30 on an active day. Background tasks such as conversation summaries use very little, typically under $0.04 per session. Developers who rely on Claude Code daily will find tips on keeping those figures down in our guide to the best ways to use Claude Code.

How to track token usage on the Claude API

Per-request usage data

Every response from the Claude API includes a usage object that lists input tokens, output tokens, tokens written to the cache and tokens read from it. Logging these fields per request gives the most precise picture of where money goes.

A separate token counting endpoint estimates input size before sending a request, which helps with budgeting large prompts. Remember that thinking tokens are billed at the output rate even when the visible answer is short.

The Claude Console usage and cost pages

The Usage page in the Claude Console charts input and output tokens over time, filtered by model, date, API key and workspace. It also shows rate-limit charts for input and output tokens per minute, including the share served from cache, and highlights requests that hit rate limits. Data can be exported as CSV.

The Cost page shows spending by model and month, including charges for web search and code execution. Access depends on Console role: Developer, Billing and Admin roles can view it. The Cost page does not break spending down by individual user, so teams that need per-person figures should issue separate API keys or workspaces.

The usage and cost admin API

For automated reporting, Anthropic offers a Usage and Cost Admin API that returns the same data programmatically. It requires an admin API key, which starts with sk-ant-admin, created by an organization administrator.

Teams that connect Claude Code to an API account will also find a dedicated Claude Code workspace created automatically, where they can set spend limits.

Usage tracking for team and enterprise plans

Administrators on Team and Enterprise plans can open a Claude Code analytics dashboard in the Claude admin area to see adoption and usage across their organization. Spend reports can be downloaded as CSV files. Enterprise customers can also use the Enterprise Analytics API and the Claude Code Analytics API to feed usage into their own dashboards.

Companies running Claude through Amazon Bedrock or Google Vertex AI track spend in those providers’ billing tools, or route traffic through OpenTelemetry or an LLM gateway for detailed records.

Why your Claude token usage climbs faster than expected

Several factors push token counts up. Every turn in a conversation re-sends the earlier context, so long chats grow more expensive with each message. Extended thinking adds output tokens that never appear in the final answer.

Connected tools and MCP servers add their definitions to the context. Large instruction files load into every session. Newer Claude models also use a tokenizer that counts roughly 30% more tokens for the same text than older generations, so a workload that fit easily on an older model may consume noticeably more of a limit today. Our explainer on tokens as the currency of generative AI covers how counting works.

How to reduce Claude token usage

Start fresh conversations for new tasks, and in Claude Code use /clear between unrelated jobs and /compact to summarize long sessions. Pick the smallest model that handles the job, since the Claude lineup ranges widely in cost; our comparison of Claude Sonnet 5 and other Anthropic models shows where each fits.

Lower the effort setting for routine work. Keep CLAUDE.md instruction files short, ideally under 200 lines, and move specialized instructions into skills that load only when needed, a feature explained in our guide on recording repeatable skills in Claude.

Use hooks and subagents to keep bulky output out of the main conversation, and disconnect MCP servers you are not using. Clear, focused requests help too, and our guide to optimal prompt lengths explains how to size them.

If you are interested in this topic, we suggest you check our articles:

Sources: Claude Code Docs – Manage costs, Claude Help Center – Cost and usage reporting in Console, Claude Platform Docs – Usage and Cost API, Claude Monitor

Written by Alius Noreika