Extra Usage Claude: How to Track, Control, and Optimize Your Spend
Extra Usage Claude: How to Track, Control, and Optimize Your Spend
Extra usage on Claude is Anthropic's prepaid credit system that kicks in when your monthly plan allowance runs out or when third-party apps like Cursor and Claude Code need tokens. If you have ever been surprised by a depleted balance or wondered where your credits went, this guide walks through the mechanics, the tracking options, and concrete strategies for keeping costs predictable.
How Extra Usage Works in 2026
Claude subscriptions (Pro at $20/month, Team at $30/seat/month, Max at $100 or $200/month) include a monthly token budget for direct use on claude.ai and the mobile app. Extra usage is a separate, prepaid pool that serves two purposes:
- Overflow: when your monthly allowance runs out, extra usage covers the remaining requests until your next billing cycle
- Third-party apps: tools like Claude Code, Cursor, Windsurf, and Cline draw exclusively from extra usage, never from your plan allowance
This separation means that heavy Claude Code users can burn through extra usage while their plan allowance sits untouched.
Important change in 2026
Anthropic moved all third-party app billing to extra usage in early 2026. Previously, these apps shared your plan limits. Now they use a separate balance, so you need extra usage credits loaded even if your plan has tokens remaining.
Extra Usage Pricing by Model and Plan
The cost of extra usage depends on which Claude model processes your request. Here is the current token pricing:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Typical message cost | Relative cost | |---|---|---|---|---| | Haiku 4.5 | $0.80 | $4.00 | $0.002 - $0.01 | 1x (baseline) | | Sonnet 4.6 | $3.00 | $15.00 | $0.01 - $0.05 | ~4x Haiku | | Opus 4.6 | $15.00 | $75.00 | $0.05 - $0.30 | ~19x Haiku |
Agent workflows (Claude Code, Cursor Agent mode) are the most expensive because they send many requests per task, each carrying a growing context window. A single Claude Code coding session can consume $2 to $15 depending on task complexity and model choice.
How to Check Your Extra Usage Balance
On claude.ai
Go to claude.ai/settings/usage. The dashboard shows:
- Current balance: your remaining prepaid credits
- Per-app breakdown: which tools are consuming your credits (Claude Code, Cursor, etc.)
- Usage history: a timeline of credit consumption over recent billing periods
- Auto-reload settings: threshold, reload amount, and monthly cap
From your macOS menu bar with Fazm
Fazm is an open source macOS AI agent that shows your Claude usage directly in the menu bar. Instead of switching to the browser to check your balance, you can see your extra usage credits at a glance while working in any app.
Fazm connects to your Anthropic account and surfaces:
- Real-time extra usage balance
- Session cost tracking as you use Claude Code or other tools
- Spend rate so you can estimate when credits will run out
This is especially useful during long coding sessions where Claude Code is consuming credits in the background.
Setting Up Spend Controls
Auto-reload configuration
Auto-reload prevents your extra usage balance from hitting zero, which would block all third-party app requests:
| Setting | Purpose | Recommended for light use | Recommended for heavy use | |---|---|---|---| | Threshold | Balance that triggers reload | $2 | $10 | | Reload amount | Credits added per trigger | $10 | $50 | | Monthly cap | Maximum auto-spend per month | $30 | $200 |
Monthly spend limit
Set a monthly cap that matches your budget. Without a cap, auto-reload will keep adding credits as long as you keep using tokens. For teams, have an admin set organization-level limits to prevent any single user from overspending.
Per-app auditing
Review connected apps regularly. Common sources of unexpected spend:
- Cursor auto-complete: sends requests on every keystroke if enabled. Switch to manual completion or a local model for routine code completion.
- Claude Code background agents: long-running agent tasks can consume $5 to $20 per session. Monitor active sessions and terminate ones that are not making progress.
- Forgotten integrations: apps you set up weeks ago but no longer use may still be drawing credits.
Five Strategies to Reduce Extra Usage Costs
1. Match the model to the task
Not every request needs Opus. Use this decision framework:
| Task type | Recommended model | Why | |---|---|---| | Quick lookups, formatting, simple edits | Haiku 4.5 | 19x cheaper than Opus, handles routine tasks well | | Code generation, analysis, writing | Sonnet 4.6 | Good balance of capability and cost | | Complex reasoning, architecture decisions | Opus 4.6 | Worth the premium for tasks that require deep thinking |
In Claude Code, you can switch models per task with /model. In Cursor, configure the default model in settings.
2. Keep context windows small
Every message in a conversation includes the full history. A 50-message thread costs significantly more per message than a fresh conversation. Start new chats for new topics rather than continuing long threads.
3. Use plan allowance first
For tasks that can run on claude.ai directly (writing, research, analysis), use the web interface before reaching for Claude Code. This draws from your plan allowance, which is included in your subscription and resets monthly.
4. Set up spending alerts
Configure auto-reload with a monthly cap. When you approach the cap, Anthropic sends a notification. This acts as an early warning before you hit unexpected costs.
5. Monitor with Fazm
Install Fazm for real-time visibility into your extra usage balance from the macOS menu bar. Seeing your spend in real time changes behavior, just like watching your electricity meter while running the AC.
Extra Usage Across Claude Plans
Each plan handles extra usage slightly differently:
| Plan | Monthly cost | Plan allowance | Extra usage available | Third-party app support | |---|---|---|---|---| | Free | $0 | Limited | No | No | | Pro | $20/month | Standard | Yes, prepaid | Yes | | Team | $30/seat/month | Higher | Yes, admin-controlled | Yes | | Max 5x | $100/month | 5x Pro | Yes, prepaid | Yes | | Max 20x | $200/month | 20x Pro | Yes, prepaid | Yes |
Max plans include significantly more tokens in the plan allowance, which means you may need less extra usage for overflow. However, third-party apps still draw exclusively from extra usage regardless of plan tier.
Tip for Max subscribers
If you are on Max 20x and primarily use Claude Code, most of your spend will still come from extra usage since Claude Code is a third-party app. The 20x plan allowance helps with direct claude.ai usage but does not reduce your extra usage consumption from developer tools.
Common Questions
Do extra usage credits expire? No. Once purchased, credits remain in your account until consumed. They do not reset or expire at the end of a billing cycle.
Can I get a refund on extra usage credits? Anthropic does not offer refunds on purchased extra usage credits. Start with a small amount ($5 to $10) to calibrate your burn rate before loading larger amounts.
Why did my extra usage balance drop overnight? Check for background processes. Claude Code sessions, Cursor auto-complete, and other integrations can consume credits while you are not actively watching. Review the per-app breakdown at claude.ai/settings/usage.
What happens when extra usage hits $0? All third-party app requests are immediately blocked. Direct claude.ai usage continues if your plan allowance has tokens remaining. If both are exhausted, all requests are blocked until your next billing cycle or you add more credits.
Is extra usage the same as the API? No. Extra usage is for consumer products (Claude Pro, Team, Max). The API has its own billing system with separate pricing. If you are building applications, use the API directly rather than routing through extra usage.
Wrapping Up
Extra usage on Claude is the billing layer that powers third-party integrations and prevents interruptions when your plan runs out. The key to managing it well: track per-app consumption, match models to tasks (Haiku for simple work, Opus for complex reasoning), and set monthly spending caps that match your budget.
Fazm is an open source macOS AI agent that tracks your Claude usage in the menu bar. Start with free built-in credits for desktop automation, then connect your own Claude subscription when ready. Open source on GitHub.