Claude Code Pricing Guide: Token Costs, API Tiers, and Subscription Plans (2026)
Anthropic’s Claude Code CLI has established itself as the premier terminal-native agentic coding tool for software engineers. Unlike traditional chat interfaces, Claude Code inspects local repositories, edits files, executes bash tests, and handles git commits autonomously.
Understanding how Claude Code is billed across Claude Pro/Team flat subscriptions versus Anthropic Console pay-as-you-go API keys is crucial for engineering teams looking to control monthly AI tooling expenditures.
Executive Summary & Plan Comparison
Claude Code offers two authentication pathways: OAuth web subscription login or Console API billing.
┌────────────────────────────────────────────────────────────────────────┐
│ CLAUDE CODE ACCESS & BILLING │
└───────────────────────────────────┬────────────────────────────────────┘
│
┌─────────────────────────┴─────────────────────────┐
▼ ▼
┌──────────────────────────────┐ ┌──────────────────────────────┐
│ SUBSCRIPTION PLANS (PRO/TEAM)│ │ CONSOLE API PAY-AS-YOU-GO │
├──────────────────────────────┤ ├──────────────────────────────┤
│ • Flat Rate: $20 - $30/mo │ │ • Per-Token Usage Billing │
│ • Fixed Hourly Usage Caps │ │ • Higher Rate Limit Tiers │
│ • Best for Individual Devs │ │ • Best for CI/CD & Large Repos│
└──────────────────────────────┘ └──────────────────────────────┘| Tier / Plan | Monthly Cost | Authentication Method | Usage Limits | Best For |
|---|---|---|---|---|
| Claude Pro | $20 / month | OAuth Login (claude login) | ~45 msgs / 5 hrs (dynamic) | Individual developers |
| Claude Team | $30 / user / month (min 5) | Team SSO OAuth | 5x Pro usage allowance | Engineering squads |
| Claude Enterprise | Custom contract | SSO & SCIM Directory | Maximum pooled quota | Enterprise governance |
| Console API (Tier 1-4) | Pay-as-you-go | API Key (ANTHROPIC_API_KEY) | Up to 10M Tokens/min | High-volume pipelines & CI/CD |
Model Token Rates (Anthropic Console API)
When operating Claude Code via an Anthropic Console API key, token consumption is billed per token processed across base input, prompt caching, and generation output:
| Model | Base Input Tokens | Prompt Cache Write | Prompt Cache Read | Output Tokens |
|---|---|---|---|---|
| Claude 3.7 Sonnet | $3.00 / 1M | $3.75 / 1M | $0.30 / 1M (-90%) | $15.00 / 1M |
| Claude 3.5 Sonnet | $3.00 / 1M | $3.75 / 1M | $0.30 / 1M (-90%) | $15.00 / 1M |
| Claude 3.5 Haiku | $0.80 / 1M | $1.00 / 1M | $0.08 / 1M (-90%) | $4.00 / 1M |
| Claude 3 Opus | $15.00 / 1M | $18.75 / 1M | $1.50 / 1M (-90%) | $75.00 / 1M |
[!TIP] Claude 3.7 Sonnet supports an optional “Extended Thinking” mode. Thinking tokens are billed as standard output tokens ($15.00 / 1M), so setting strict reasoning token budgets helps keep complex debugging loops economical.
4 Strategies to Cut Claude Code API Costs by 80%
1. Optimize CLAUDE.md Architecture
Large system instruction files get sent with every user prompt. Keep CLAUDE.md under 150 lines by keeping only essential project build commands and style conventions. Trimming redundant guidelines saves thousands of input tokens per interaction.
2. Maximize Prompt Caching
Claude Code automatically applies Anthropic’s prompt cache headers (5-minute TTL). Consecutive prompts sent within the cache window read the repository context at $0.30 per 1M tokens instead of $3.00, yielding immediate 90% savings.
3. Use /compact Proactively
As Claude Code runs terminal commands and views files, session context expands. Running /compact strips stale compiler errors and shell outputs, resetting the active context window to a lightweight, cheap footprint.
4. Switch to Haiku for Routine Scripting
For simple refactoring tasks or single-function unit test generation, setting the environment model to Claude 3.5 Haiku reduces token expenses by 73% relative to Sonnet.
