Claude surfaces map

Every surface, how they relate, and where your tokens go — Leslie Tou · March 2026

Products

Click any product to highlight its surfaces and billing. Click again to clear.

Claude

Conversational interface. The default Claude experience — chat, projects, files, Deep Research.

All plans200K–500K contextBrowser · Desktop · Mobile

Claude Code

Agentic coding tool that reads your real codebase, edits files, and runs commands.

Pro / MaxAPI key opt-inCLI · VS Code · Desktop · Web

Cowork

Claude Code capabilities for non-developers. File, folder, and document tasks. Desktop only.

BetaDesktop onlyMobile Remote

Claude API

Raw programmatic access. Build products on Claude. Pay per token, no usage caps.

Pay-per-tokenUp to 1M contextAny language

Specialty agents

Task-specific agents: Deep Research, Chrome, Excel, PowerPoint. Token-heavy; in beta.

BetaPlan usageDeep Research · Chrome · Excel · PPT

Also available via third partiesCopilot, Cursor, Amazon Bedrock, Google Vertex AI, Microsoft Foundry — Claude model, their contract.

Surface map

Organized by where you actually open it — not by product category.

Claude desktop app

Chat

Claude· Synced

History synced server-side with claude.ai and mobile.

Code

Claude Code· Local

GUI wrapper around Claude Code. History stored locally.

Cowork

Cowork· Local

File and document tasks. Stored locally, not subject to Anthropic retention.

Three tabs, one app. All count against your plan usage pool.

Browser

claude.ai

Claude· Synced

History synced across all devices. Start here, continue anywhere.

Deep Research

Agent· Synced

Long-horizon research with structured reports. Runs inside claude.ai.

Claude in Chrome

Agent· Synced

Browses live web, fills forms. History synced via claude.ai — shows up across devices.

Claude Code (web)

Claude Code· Remote

Remote Claude Code sessions. History on Anthropic servers — separate from local CLI history.

Mobile

Claude (iOS / Android)

Claude· Synced

Full sync with claude.ai. Start a chat anywhere, pick it up anywhere.

Dispatch

Cowork· Synced

Assign and monitor Cowork tasks from your phone. Task executes locally on your desktop — phone is the control surface.

Remote Control

Claude Code· Remote

Run /rc in the CLI to get a URL. Open it on your phone to see Claude's edits, approve changes, and send follow-ups. Max plan only, research preview.

Dispatch and Remote Control don't move execution to the cloud — your local machine stays in charge.

Terminal & editor

Claude Code CLI

Claude Code· Local

Full agentic coding in your terminal. Run: claude

VS Code — Anthropic plugin

Claude Code· Local

Claude Code inside your editor. Same local history as the CLI. Not the same as Copilot chat.

CLI and VS Code plugin share the same local history at ~/.claude/history.jsonl

Microsoft Office

Claude in Excel

Agent· No history

Formula help, data analysis, bulk edits. The spreadsheet holds the output.

Claude in PowerPoint

Agent· No history

Slide generation and editing. The deck holds the output.

No chat history — each session starts fresh. The file is the memory.

API & cloud

Cursor (BYOK)

API

Cursor with your Anthropic key. Hits your API account, not your plan.

OpenClaw

API· No history

Third-party Claude client using your Anthropic API key. Hits your API account — not your claude.ai plan.

Third-party

VS Code — Copilot chat

Third-party

Haiku free on GitHub free tier. Sonnet/Opus cost Copilot premium requests. No Anthropic account needed.

Cursor (default)

Third-party

Cursor's own subscription. Not billed to Anthropic.

Claude model, someone else's contract. Doesn't touch your Anthropic account.

Billing

claude.ai plan

Pro / Max — flat monthly fee

claude.ai (browser, mobile, desktop Chat)

Deep Research

Desktop Code + Cowork tabs

VS Code Claude plugin

Claude Code CLI + browser

Chrome, Excel, PowerPoint agents

All share one usage pool. Cowork, Chrome, and Code eat from the same limit as chat.

Anthropic API account

Pay-per-token — for building products on Claude

Cursor with your API key (BYOK)

OpenClaw (uses your API key)

Batch API jobs

Any app you've given your key

Direct API calls

Additive — runs alongside your plan. Two Anthropic bills can stack. Neither dashboard shows the other.

Third-party billing

Not on your Anthropic account

VS Code Copilot chat (GitHub/Microsoft)

Cursor without BYOK

Amazon Bedrock (billed via AWS)

Google Vertex AI (billed via GCP)

Microsoft Foundry (billed via Microsoft)

Claude model, someone else's contract.

Plan tiers & usage limits

Limit	Free	Pro $20/mo	Max 5× $100/mo	Max 20× $200/mo
Session limit resets every 5 hrs	1× baseline	5× vs Free	25× vs Free	100× vs Free
Weekly limit resets 7 days after window opens	1× baseline	5× vs Free	25× vs Free	100× vs Free
Sonnet weekly meter Sonnet-only cap	Hard cap	Soft cap. Hit it → Sonnet blocked, Opus and Haiku still available. Extra usage unlocks it if on.	Soft cap. Hit it → Sonnet blocked, Opus and Haiku still available. Extra usage unlocks it if on.	Soft cap. Hit it → Sonnet blocked, Opus and Haiku still available. Extra usage unlocks it if on.
All-models weekly meter Opus + everything else	Hard cap	Soft cap. Hit it → everything blocked. Extra usage unlocks it if on. Real gate: your monthly spending cap.	Soft cap. Hit it → everything blocked. Extra usage unlocks it if on. Real gate: your monthly spending cap.	Soft cap. Hit it → everything blocked. Extra usage unlocks it if on. Real gate: your monthly spending cap.
Extra usage pay-as-you-go above limit	Not available	Yes, at API rates	Yes, same rates	Yes, same rates
At limit	Wait or upgrade	Wait, or extra usage	Wait, or extra usage	Rarely hit

All surfaces share one pool. A heavy Cowork session, Claude Code marathon, and regular chat all drain the same session and weekly limits.

Extra usage is additive billing — it appears as a separate charge on top of your subscription.

Weekly limits are the real wall. Session limits reset in 5 hours. Weekly limits don't. Blow through your weekly on Monday and you're stuck until reset.

The spending cap can lock you out. You can have credits in the account and still be blocked if your monthly cap says stop. Auto-reload is off by default.

Sonnet double-counts (known bug). On Pro/Max, Sonnet usage drains both the Sonnet meter and the all-models meter simultaneously — so you may hit the all-models cap faster than expected even if you're only using Sonnet.

Off-peak promotions exist. Anthropic occasionally doubles limits during off-peak hours, and that bonus usage doesn't count toward weekly caps.