Is Claude Code free? No — here is the cheapest way in, and the agents that start at $0

Claude Code is not free. Anthropic's pricing page lists it as not included on the Free plan, and every way to run it costs money: a paid Claude plan, or an API key billed per token. What is free is a handful of other coding agents — and a tool that shows what each of them costs you.

Is Claude Code free?

No. Claude Code comes with Pro, Max, Team and Enterprise, not with Claude Free. Installing it costs nothing, but it will not run without a paid plan or an API key behind it. The free credits Anthropic gives new API accounts are, in its words, a small amount to test the API; it does not publish how much.

Claude Code free tier

There is none. Claude Free gets chat in claude.ai, with its own five-hour limit; Claude Code starts at Pro. Claude in another agent, or through a cloud provider, is billed per token too.

The cheapest way to run Claude Code

Two ways in, and which is cheaper depends on how much you use it.

Claude Pro. $20 a month, or $17 a month billed annually, which is $200 up front. Claude Code shares Pro's allowance with claude.ai: a five-hour window and a weekly limit. Past them, you wait, buy usage credits, or switch to an API key.

An API key. No plan and no plan limits; every token is billed at the list price of the model you pick. Haiku 5.5 is the cheapest current model: $0.10 per million input tokens and $0.50 per million output for a prompt up to 100,000 tokens, $0.50 and $2.50 above that — and an agent's prompt passes 100,000 tokens quickly. Sonnet 5.5 and Sonnet 5 are $2 and $10. An agent re-sends the whole conversation on every call, mostly from the cache, so the bill grows with the length of a session as well as the model: my median day, mostly on Opus, cost $153 at list price — more than seven months of Pro.

Not the Batch API. Anthropic's Batch API halves the price, but it runs requests asynchronously — most batches finish within the hour — and an agent needs each reply before it can take the next step. Claude Code sends every call live, at the standard rate.

What a month of Claude Code costs

Anthropic's own figure, across enterprise deployments, is about $13 per developer per active day and $150 to $250 a month, with 90% of users under $30 a day. My own use is far heavier: at list price, my median active day came to $153, and half my days fell between $75 and $373. Those are list prices; on a plan, I paid the fee, and the limits decide how far it goes: see Claude usage limits.

The plan arithmetic is in Claude Code pricing, with a calculator for your own days.

Coding agents that start at $0

If $0 is the requirement, the agent has to change, and usually the model with it.

Antigravity. Google's agent — a desktop app, an IDE and a CLI, agy — is free on a Google account with a weekly quota. It replaced Gemini CLI for free users on 18 June 2026; Gemini CLI itself now needs a Code Assist licence or a paid API key. Antigravity vs Claude Code · Gemini CLI vs Claude Code.

OpenCode. Free and open source, MIT, and it connects to more than 75 model providers. The agent costs nothing; the model is billed by whoever serves it. OpenCode's own Zen service listed 12 models as free on 3 October 2026, each for a limited time, and on several of them data collected in the free period may be used to improve the model. On a cheap open model the median call on my machine cost $0.002. A Claude Pro or Max sign-in does not work in OpenCode; Claude there is billed per token. OpenCode vs Claude Code.

Codex on ChatGPT Free. OpenAI lists Codex on ChatGPT Free and Go in the desktop app, subject to rollout, on its GPT-6 Luna model. Plus at $20 opens the CLI, web and IDE extension. Codex vs Claude Code.

Cursor Free. Cursor's free plan includes limited use of its own models, Grok and Composer, in its own editor. Claude in Cursor starts at Premium, $30 a month, billed at API list price on top.

Every Claude Code alternative, free and paid, on price, limits, models and licence: AI coding agents compared.

See what each one costs you

Free agents still have a cost — your time on the limits, or a provider's bill — and paid ones have a list price behind the flat fee. Caprock is free and open source (Apache-2.0) and runs on your machine. It reads Claude Code's and Codex's transcripts and OpenCode's database, whether or not it started the session, and Gemini CLI in the sessions it starts, and prices every call, so you can see what a day of each costs before you pick a plan. The dollars are a list-price estimate, as Claude Code's own /usage is, beside the plan's dollars and the share of the weekly limit the work used. It does not read Cursor.

Nothing leaves your machine. Install Caprock and it reads back every session those agents still have on disk.