Plans at a Glance
Free
$0
via claude.ai
- Limited Claude Code usage
- Rate limits apply
- Good for trying Claude Code
- Not for daily development
API (Usage-based)
~$10–40
per month (typical dev)
- Pay per token used
- Full model access
- No rate limits (soft quotas)
- Most flexible option
Claude.ai Pro / Max
$20–100
per month (flat rate)
- Higher usage caps
- Predictable billing
- Access via claude.ai/code
- Best for regular heavy users
Token Pricing (Anthropic API)
Claude Code uses the Anthropic API when accessed via API key. You pay per million tokens processed:
| Model | Input ($/M tokens) | Output ($/M tokens) | Cache Read ($/M) | Cache Write ($/M) |
|---|---|---|---|---|
| Claude Sonnet 4.6 (default) | $3.00 | $15.00 | $0.30 | $3.75 |
| Claude Opus 4.7 (most capable) | $15.00 | $75.00 | $1.50 | $18.75 |
| Claude Haiku 4.5 (fastest/cheapest) | $0.80 | $4.00 | $0.08 | $1.00 |
Prompt caching dramatically reduces costs. Once context (your codebase, CLAUDE.md) is cached, re-reading it costs 90% less. A 100k-token codebase context costs $0.30 to cache — subsequent reads cost only $0.03.
Real-World Cost Examples
Example 1
Quick bug fix (10 min session)
~15k input tokens (code context) + ~3k output tokens (explanation + fix). No caching needed.
≈ $0.09
Example 2
Full feature implementation (1 hour)
~80k input tokens (repo context, prompts, test output) + ~20k output tokens (code, explanations). With caching after first read.
≈ $0.50–$1.20
Example 3
Large refactor across 30 files (2 hour session)
~200k input tokens (full repo + multiple iterations) + ~40k output tokens. Heavy use of prompt cache for repeated context.
≈ $1.50–$3.50
Example 4
Full /ultrareview of a large PR
~120k input tokens (diff + full context) + ~15k output tokens (detailed analysis). Single session, no caching benefit.
≈ $0.60–$1.00
Example 5
Daily developer (8 hours, mixed tasks)
Typical heavy developer day: feature work, debugging, reviews, quick questions. ~500k input tokens + ~80k output with good caching.
≈ $2.50–$5.00/day → ~$60–120/month
How to Reduce Your Claude Code Bill
Tip 1: Enable prompt caching. Claude Code uses caching automatically for CLAUDE.md and stable context. Make sure your CLAUDE.md is comprehensive — the more stable context you cache, the cheaper each subsequent message becomes.
Tip 2: Use /compact mid-session. Running
/compact compresses conversation history, reducing the tokens needed to maintain context in long sessions. Use it every 30–40 minutes during extended work.
Tip 3: Scope tasks clearly. Vague prompts lead to unnecessary exploration. Instead of "look at the auth system", say "check
src/auth/middleware.ts lines 40–80 for the session validation bug". Less token-hunting = lower cost.
Tip 4: Use Haiku for lightweight tasks. Model routing lets you run quick questions or simple edits through Claude Haiku (3.75× cheaper than Sonnet). Reserve Sonnet/Opus for complex reasoning tasks.
Tip 5: Avoid unnecessary file reads. Claude Code automatically reads files when given permission. Narrow the allowed paths in your session to avoid it reading large swaths of your repo when you only need a small section.
Claude Code vs Competitors on Price
| Tool | Pricing Model | Monthly Cost (typical dev) | Best For |
|---|---|---|---|
| Claude Code (API) | Pay-per-token | $15–50 | Complex autonomous tasks |
| GitHub Copilot | Flat $10/mo | $10 | Inline completions, daily coding |
| Cursor Pro | Flat $20/mo | $20 | AI-first IDE, completions + chat |
| OpenAI Codex / ChatGPT | Usage-based | $20–60 | Code generation, chat |
Claude Code's per-token model means cost scales with work done. A 30-minute autonomous refactor that saves 3 hours of manual work might cost $1–2. At that ROI, the cost comparison shifts — you're paying for outcomes, not seat licenses.
See What You Get for That Cost
Browse every Claude Code skill, hook, MCP server, and automation capability to understand the full value.
Open Claude Skills Browser →