What Does /cost Do in Claude Code? (Token Usage Tracking)

Q: What does /cost show you?

Type /cost and you'll see a breakdown of: - Input tokens — everything Claude has read (your messages, file contents, tool results) - Output tokens — everything Claude has written back (responses, code, explanations)

Q: How much does a typical session cost?

Real-world usage data shows the average is around $13 per developer per active day across enterprise deployments, with 90% of users staying below $30 per day. But individual sessions vary wildly: | Session Type | Typical Tokens | Estimated Cost (Sonnet) | |-------------|---------------|------------------------|

Q: When should you check /cost?

Check /cost when: - Before compacting — to see how much context you've accumulated - After a large operation — reading big files or running many tool calls can spike usage

Q: How do you keep costs down?

1. Use /compact proactively — compressing at 60% context saves tokens on every subsequent message 2. Switch models with /model — use Haiku for simple tasks, Sonnet for most work, Opus only for complex reasoning 3. Use /btw for quick questions — side questions don't accumulate in your main context

Wondering how much your Claude Code session is costing? The /cost command shows you exactly how many tokens you've used and an estimated dollar amount for your current session.

What does /cost show you?

Type /cost and you'll see a breakdown of:

Input tokens — everything Claude has read (your messages, file contents, tool results)
Output tokens — everything Claude has written back (responses, code, explanations)
Total tokens — combined usage for the session
Estimated cost — a dollar figure based on current model pricing

The dollar estimate is calculated locally from your token counts and the model's published rates. It may differ slightly from your actual bill — for authoritative billing, check the Usage page in the Claude Console.

How much does a typical session cost?

Real-world usage data shows the average is around $13 per developer per active day across enterprise deployments, with 90% of users staying below $30 per day. But individual sessions vary wildly:

Session Type	Typical Tokens	Estimated Cost (Sonnet)
Quick bug fix	20K–50K	$0.10–$0.50
Feature implementation	50K–150K	$0.50–$2.00
Large refactor	150K–300K	$2.00–$5.00
Full-day coding	300K–500K+	$5.00–$15.00

On API pricing (Sonnet 4.6: $3/1M input, $15/1M output), a session that reads 100K tokens of code and generates 20K tokens of output costs roughly $0.60.

On Pro ($20/month) or Max plans, you're paying a flat rate — /cost still shows tokens consumed, which helps you understand your usage relative to your plan limits.

When should you check /cost?

Check /cost when:

Before compacting — to see how much context you've accumulated
After a large operation — reading big files or running many tool calls can spike usage
Planning your day — if you're on API credits, knowing your burn rate helps you budget
Hitting rate limits — high token usage per window is usually why you're being throttled

How do you keep costs down?

Use /compact proactively — compressing at 60% context saves tokens on every subsequent message
Switch models with /model — use Haiku for simple tasks, Sonnet for most work, Opus only for complex reasoning
Use /btw for quick questions — side questions don't accumulate in your main context
Be specific in prompts — vague requests cause Claude to read more files and explore more options
Use .claudeignore — exclude large directories (node_modules, build artifacts) from Claude's file scanning

/cost vs external monitoring tools

/cost is built-in and shows your current session. For historical tracking across sessions, the community has built tools like:

ccusage — CLI tool that reads Claude's local JSONL logs to show usage by date, session, or project
Claude-Code-Usage-Monitor — real-time dashboard with token consumption charts and limit predictions

These complement /cost by giving you the bigger picture of your spending patterns over time.