Claude Code is not free and doesn’t have a single price. There are two billing tracks: a flat monthly subscription that bundles Claude Code access with a usage allowance, and pay-as-you-go API billing where every token you send and receive has a cost. Whether a subscription or the API is cheaper depends entirely on how many hours per day you actually code. This guide walks through exact plan rates, the per-token model costs that drive API bills, what real monthly spend looks like, and the most effective ways to reduce it.
Subscription Plans That Include Claude Code
Claude Code is included in every paid Anthropic plan — but not in the free tier. If you’re on the free plan, you can use claude.ai for chat but you cannot use the Claude Code CLI or agentic features.
| Plan | Monthly Price | Annual (per month) | Notes |
|---|---|---|---|
| Pro | $20 | ~$17 | Individual; includes Claude Code |
| Max 5x | $100 | — | 5× Pro usage allowance |
| Max 20x | $200 | — | 20× Pro usage allowance |
| Team Standard | $25/seat | ~$20/seat | Shared workspace; Claude Code access |
| Team Premium | $125/seat | ~$100/seat | Full Claude Code for engineering teams |
| Enterprise | Custom | Custom | Negotiated; includes Claude Code |
Source: claude.com/pricing, support.anthropic.com, checked July 2026.
Pro ($20/month)
Pro is the entry point. At $20/month (or about $17 billed annually), it unlocks the full model stack — Haiku 4.5, Sonnet 5, and Opus 4.8 — plus Claude Code. It’s the right choice for an individual developer who uses Claude Code regularly but not continuously throughout the day.
The caveat: Pro has a usage ceiling. Claude Code and long agentic sessions burn allowance faster than simple chat, so heavy users hit the limit. Anthropic’s /usage command shows where you stand; if you’re consistently hitting the Pro ceiling before the billing period resets, Max 5x is the logical upgrade.
Max Plans ($100–$200/month)
Max 5x at $100/month gives 5× Pro’s usage allowance. Max 20x at $200/month gives 20×. Both tiers are designed specifically for developers who run extended coding sessions, multi-agent workflows, or have Claude Code running in the background most of the workday.
At $100/month, Max 5x breaks even against the API at roughly 3–4 active coding days per week on Sonnet. If you’re using Claude Code daily and regularly, the subscription math usually wins.
Team and Enterprise
Team Standard ($25/seat/month) includes Claude Code and is a reasonable baseline for small engineering teams. Team Premium ($125/seat/month) is the practical equivalent of a Max plan per seat — use it when the team needs full Claude Code capacity with no individual usage ceilings. Enterprise adds compliance controls (SSO, audit logs, custom data handling) on top of the Team Premium feature set.
Claude Code API Token Costs
If you’re not on a subscription, or if you’ve configured Claude Code to use a separate API key, you’re on pay-as-you-go billing. Costs scale directly with how much context you load and how much the model outputs.
| Model | API ID | Input (per MTok) | Output (per MTok) | Context |
|---|---|---|---|---|
| Claude Haiku 4.5 | claude-haiku-4-5-20251001 | $1.00 | $5.00 | 200k tokens |
| Claude Sonnet 5 | claude-sonnet-5 | $2.00* | $10.00* | 1M tokens |
| Claude Opus 4.8 | claude-opus-4-8 | $5.00 | $25.00 | 1M tokens |
| Claude Fable 5 | claude-fable-5 | $10.00 | $50.00 | 1M tokens |
*Sonnet 5 introductory pricing through August 31, 2026; standard rate is $3/$15 per MTok.
Source: docs.anthropic.com, checked July 2026.
Why output tokens matter more than you think
Claude Code doesn’t just send short queries. A single agentic loop can include file reads, tool call results, and multi-step reasoning — all of which flow through the context. Output tokens are 4–5× more expensive per token than input tokens on Sonnet and Opus. On a long session where Claude is generating tests, reviewing diffs, and writing code across several files, the output bill can easily exceed the input bill.
Prompt caching and Batch API
Anthropic applies prompt caching automatically on repeated context blocks, which matters for Claude Code sessions where the same file contents get re-read across turns. Cache reads are significantly cheaper than fresh input tokens. The Batch API (for async, non-realtime workloads) cuts both input and output rates by 50%.
What Does Claude Code Actually Cost Per Month?
Across enterprise deployments, Anthropic reports an average of $13 per developer per active coding day, with 90% of users staying under $30 per active day. That works out to roughly $150–$250 per developer per month for typical daily use. Source: docs.anthropic.com/en/docs/claude-code/costs.
These are API costs, not subscription costs. The numbers are also averages — junior developers on simple tasks land lower; senior developers doing large refactors or architecture analysis on big codebases land higher.
Subscription vs. API: where the break-even is
At Sonnet 5 intro pricing ($2/$10 per MTok), $20/month buys you roughly 10 active coding days before the API spend matches the Pro subscription cost. If you’re coding with Claude Code more than 10 days per month, a Pro subscription is cheaper than raw API billing — assuming you stay within the usage allowance.
Max 5x ($100/month) break-evens at approximately 7–8 active coding days on Opus 4.8. Developers running long Opus sessions for complex work typically see the Max subscription pay off within the first week.
Five Ways to Reduce Your Claude Code Bill
1. Track usage with /usage
Type /usage inside any Claude Code session to see a token breakdown by tool, subagent, and MCP server. On subscription plans it shows where you are against your allowance. On API billing it shows the session cost estimate. Making this a habit at the end of sessions reveals where your tokens are actually going — often it’s file reads and tool calls, not the conversation itself.
2. Set a monthly spend limit
On Pro and Max plans, the /usage-credits command opens a dialog to set a hard monthly spend cap. For API users, the Claude Console has usage alerts and hard limits under the billing settings. Set a limit before you start heavy usage, not after.
3. Match the model to the task
The default model is Sonnet, and for most coding tasks that’s the right call. Switch to Haiku for simple edits, comment cleanup, or high-volume scripted runs. Haiku at $1/$5 per MTok is 2–5× cheaper than Sonnet and noticeably faster; for mechanical tasks, it’s the correct choice.
Reserve Opus for genuinely hard problems: architectural decisions, root-cause debugging, or large codebase analysis where reasoning depth matters. Opus costs 2.5× more than Sonnet per token and is slower. Don’t use it by default just because it’s the most capable.
See Claude Code Default Model: How to Change, Switch, and Choose the Right One for a full breakdown of when each model tier earns its cost.
4. Write specific prompts
“Improve this codebase” triggers broad file scanning. “Add input validation to the createUser function in src/auth/users.ts” tells Claude exactly where to look. Every file Claude reads is tokens loaded into context. Targeted prompts reduce context size and total session cost.
5. Use /compact and context management
Long sessions accumulate context quickly. The /compact slash command summarizes the conversation history into a compressed form, shedding token weight from older turns. Running /compact before starting a new subtask within a long session can meaningfully reduce costs on API billing. See Claude Code CLI Documentation for the full list of commands that affect context behavior.
Open-Source Models as a Cost Alternative
Before Novita’s Agent Plan launches, developers looking to cut costs without abandoning Claude Code’s agentic workflow have a practical option: route Claude Code’s API requests to Novita AI’s LLM API instead of the default Anthropic endpoint.
Claude Code reads the ANTHROPIC_BASE_URL environment variable. Point it at Novita AI’s Anthropic-compatible endpoint and it will send requests to Novita-hosted models instead. The setup is three environment variables:
export ANTHROPIC_BASE_URL="https://api.novita.ai/anthropic"
export ANTHROPIC_AUTH_TOKEN="<your Novita API key>"
export ANTHROPIC_MODEL="meta-llama/llama-3.1-70b-instruct"Novita AI hosts 200+ open-weight models. Several are strong performers for coding work:
| Model | Novita Input (per MTok) | Novita Output (per MTok) | Notes |
|---|---|---|---|
| DeepSeek V4 Pro | $1.60 | $3.20 | Strong code and reasoning |
| Qwen3 Coder 30B A3B | $0.07 | $0.27 | Code-optimized MoE |
| Llama 3.1 8B Instruct | $0.02 | ~$0.02 | Ultra-low cost, simple tasks |
Source: novita.ai/models/llm, checked July 2026.
Compared to Claude Sonnet 5 at $2/$10, DeepSeek V4 Pro at $1.60/$3.20 cuts output costs by 68%. Qwen3 Coder 30B A3B at $0.07/$0.27 is an even more aggressive option for high-volume code editing workloads. For teams running parallel agents, that difference accumulates fast.
The trade-off is honest: open-source models don’t support Claude-specific features like extended thinking, native tool calling in Claude’s format, or Anthropic’s safety classifiers. For code-focused agentic tasks — file editing, test generation, refactoring — the capability gap is smaller than in reasoning-heavy work. If you’re doing architecture analysis or complex multi-step debugging, the Claude models earn their higher cost. If you’re doing repetitive code edits or documentation passes, an open-source alternative often performs well enough at a fraction of the price.
Start with free trial credits on novita.ai, test your actual workload, and measure before committing.
FAQ
Does Claude Code have a free tier?
No. The free Claude plan covers claude.ai chat only. Claude Code — the CLI, VS Code extension, and agentic capabilities — requires Pro ($20/month) or above. Source: support.anthropic.com.
How much does Claude Code cost per month for a typical developer?
Enterprise-level data from Anthropic puts average API cost at around $13 per active coding day and $150–$250 per developer per month. On a subscription, Pro at $20/month covers moderate daily use; Max 5x at $100 handles heavy daily coding.
When is API billing cheaper than a subscription?
If you use Claude Code fewer than ~10 days per month on Sonnet (at current intro pricing), raw API billing is roughly equivalent to or cheaper than the $20 Pro subscription — assuming you’re not hitting Pro’s usage ceiling. For daily users, a subscription plan is almost always more cost-effective.
What is the cheapest way to use Claude Code?
For subscription users: Haiku 4.5 is the lowest-cost model, use /compact regularly, and write targeted prompts to minimize file reads. For API users: Haiku 4.5 at $1/$5 per MTok is cheapest within Claude’s model family. For open-source alternatives, Novita AI’s Llama 3.1 8B at ~$0.02/MTok is the floor.
Can I set a spending limit on Claude Code?
Yes. Subscription users can set a monthly spend limit via the /usage-credits command in the CLI. API users can configure hard spend limits and usage alerts in the Claude Console.
Source: Anthropic docs — manage costs, claude.com/pricing, novita.ai/models/llm. Prices checked July 15, 2026.
Novita AI is an all-in-one cloud platform for AI development. Access 200+ LLM models via a single API, deploy on serverless infrastructure, and run GPU instances — with pricing designed to keep costs predictable at any scale.
