Is Claude Code free? Every free way to run a terminal coding agent

What Claude Code costs now, who gets it free, and every other way to run a terminal coding agent for $0: free plans, open source agents with free or cheap models, and local models.

Updated October 2026

No, Claude Code isn't free. Anthropic's setup page says "the free claude.ai plan does not include Claude Code access", and the cheapest way in is Claude Pro at $20 a month. The one exception is Anthropic's program for open source maintainers. If you want a terminal coding agent for $0, there are real options: Codex on ChatGPT Free, GitHub Copilot CLI on Copilot Free, Google's Antigravity CLI, open source agents on free or local models, and Darce's free daily allowance.

Last updated: October 2026. This page is published by the makers of Darce. Prices and limits come from each vendor's own pricing page or docs, fetched on October 15, 2026. Free tiers change often, so check the linked source before you rely on one.

What Claude Code costs now

Claude Code is included in every paid Claude plan, and in none of the free ones.

PlanPriceClaude Code
Free$0Not included
Pro$20/month, or $17/month billed yearlyIncluded
Max 5x$100/monthIncluded, 5x Pro's usage
Max 20x$200/monthIncluded, 20x Pro's usage
TeamStandard seats $25/month ($20 billed yearly); Premium seats $125 ($100 billed yearly)Included with every seat
Enterprise$20/seat/month billed yearly, 20-seat minimum, plus usage at API ratesIncluded

Pro and Max have a five-hour session limit and a weekly limit, shared between the Claude apps and Claude Code. Claude's pricing page lists no free trial.

You can also skip the subscription and pay per token through a Console account. The cheapest current model, Claude Haiku 5.5, costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100K tokens. That isn't free, but light use can cost very little. Anthropic's cost docs put enterprise usage at about $13 per developer per active day, so heavy use adds up fast.

Free Claude Code for open source maintainers

Claude for Open Source gives "6 months of free Claude Max 20x" to eligible maintainers and contributors. You qualify with any one of these, among others:

  • a project with 500 or more dependent repos, 100 or more dependent packages, or 200,000 or more combined monthly downloads
  • 100 or more pull requests merged into repos you don't own in the last 12 months
  • a repo with 20 or more unique external contributors in the last 12 months

When the six months end, the account goes back to the free plan, or to your previous paid plan if you had one.

Free tiers of other coding agents

These are first-party agents with a $0 plan. Each one limits you to its own vendor's models.

AgentFree planWhat you getCatch
Codex (OpenAI)ChatGPT FreeCodex "on quick coding tasks"No published limits; the models page lists GPT-6 Luna for Free, not GPT-6.1 Sol
GitHub Copilot CLICopilot Free2,000 completions a month and an allowance of GitHub AI CreditsModels through auto selection only
Antigravity CLI (Google)Individual, $0/month"Basic weekly rate limits"Pricing page doesn't say how the CLI counts against them

Codex. OpenAI's pricing page says Codex is included in "your ChatGPT Free, Go, Plus, Pro, Business, Edu, or Enterprise plan". Free is for exploring Codex "on quick coding tasks", and OpenAI's models page says Free and Go get GPT-6 Luna (subject to rollout), not the default GPT-6.1 Sol. Go, at $8 a month, is pitched at "lightweight coding tasks". The CLI is open source, so it also runs on your own API key or a local model. See Codex vs Claude Code for how the two compare.

GitHub Copilot. GitHub's plans page says "All plans include Copilot CLI", Free included. Copilot Free is "limited to 2000 completions per month" and comes with an allowance of GitHub AI Credits; GitHub doesn't give the number on that page. On Free, you get models through auto model selection only. Paid plans are $10 (Pro), $39 (Pro+) and $100 (Max) a month.

Antigravity CLI, which replaced Gemini CLI for individuals. Google announced on May 19, 2026 that Gemini CLI is moving to Antigravity CLI. On June 18, 2026, Gemini CLI stopped serving Google AI Pro and Ultra users and people using it free of charge through Gemini Code Assist for individuals. It still works with paid Gemini API keys. Google says Antigravity CLI "is available to everyone" and signs in with a personal Google account, and Antigravity's pricing page lists a $0 individual plan with "basic weekly rate limits". It doesn't spell out the limits for the CLI.

Free tiers that ended. Qwen Code's free Qwen OAuth tier was discontinued on April 15, 2026. You now need an Alibaba ModelStudio plan, an API key or another provider. Gemini CLI's free sign-in, above, ended in June.

Open source agents with your own keys

Most open source agents are free to install. You pay whoever runs the model, which can be a free model, a cheap one, or a subscription you already have.

AgentLicenseFree to use?Ways to get free or cheap models
OpenCodeMITYes, open sourceOpenCode Zen lists a few models as free "for a limited time"; local models; ChatGPT or Copilot login
Codex CLIApache-2.0Yes, with a ChatGPT plan or keyChatGPT Free; --oss for local models
ClineApache-2.0"Free for individual developers"Your own keys; you pay for inference
Kilo CodeMIT"Free & Open Source""Auto Free" routes to free models through Kilo Gateway where available
AmpProprietaryHobby plan is freeUse your own keys or subscriptions; "No Amp token fees or limits"
AiderApache-2.0YesYour own keys or local models
GooseApache-2.0Yes15+ providers, including Ollama
CrushFSL-1.1-MITYesLocal models; Charm's Hyper has a free tier

OpenRouter's free models work with any agent that takes an OpenRouter key. Models whose ID ends in :free are limited to 20 requests a minute and 50 a day if you've bought fewer than 10 credits, or 1,000 a day once you have. An agent uses several requests per task, so 50 a day goes quickly.

Cheap paid models can be close to free for small tasks. Claude Haiku 5.5 and GPT-6 Luna are both $0.10 per million input tokens and $0.50 per million output tokens on their makers' APIs.

For a wider comparison of these tools, see best CLI coding agents and OpenCode vs Claude Code.

Local models: free if you have the hardware

Running a model on your own machine costs nothing per token. Agent work needs a model that handles tool calls well and a large context window; Ollama's docs recommend at least 64k tokens for both Claude Code and Codex.

  • Claude Code with Ollama. Ollama documents ollama launch claude, or setting ANTHROPIC_BASE_URL=http://localhost:11434 with ANTHROPIC_AUTH_TOKEN=ollama. Anthropic doesn't support Claude Code with non-Claude models, so expect rough edges. More in Claude Code with other models.
  • Codex with Ollama or LM Studio. codex --oss uses a local provider; pick one with --local-provider ollama or lmstudio. Ollama also documents ollama launch codex.
  • OpenCode has documented setups for Ollama, LM Studio and llama.cpp.

The tradeoff is hardware: you need a machine that can run a model with reliable tool calling and a 64k context at a speed you can work with.

Darce's free tier

Darce is our open source (MIT) terminal agent, so weigh this section accordingly.

  • Try it with no account. npx darce-cli gives you a short free trial with no sign-up, no API key and no card.
  • Free plan. darce signup creates a free account with a daily allowance that resets every 24 hours. It covers fast, low-cost models such as Claude Haiku, Qwen, DeepSeek and Gemini Flash. Usage is measured by what the work actually costs, so these models go a long way.
  • Every feature on every plan. The free plan has /undo for everything the agent did, including what shell commands changed, plus /derby, /swarm and /taste.
  • No surprise bills. When the allowance runs out, Darce stops; nothing is charged.
  • Or bring your own OpenRouter key, and requests are billed to your OpenRouter account instead.

What you don't get for free: Claude Sonnet and Opus, GPT and Gemini Pro start on the Builder plan at $15 a month, and Power at $65 has every model. Darce doesn't run local models. See pricing and choosing models.

Which free option should you use?

  • You want Claude Code specifically: it isn't free unless you qualify as an open source maintainer. Pro at $20 is the cheapest plan.
  • You want a first-party agent for $0: Codex on ChatGPT Free, or Copilot CLI on Copilot Free.
  • You use Google's models: Antigravity CLI's free individual plan.
  • You have a capable GPU or a Mac with plenty of memory: OpenCode, Codex or Claude Code with a local model.
  • You want many models without keys, and undo for shell commands: Darce's free plan.

Sources

Frequently asked questions

Is Claude Code free?

No. Anthropic says the free claude.ai plan doesn't include Claude Code. The cheapest plan with it is Pro at $20 a month, or $17 a month billed yearly. You can also pay per token through a Console account.

Can I get Claude Code for free as an open source maintainer?

Yes, if you qualify. Claude for Open Source gives six months of Claude Max 20x free to maintainers and contributors who meet one of its criteria, such as 500 or more dependent repos, 200,000 monthly downloads, or 100 merged pull requests to other people's repos in the last year.

What is the best free alternative to Claude Code?

For a first-party agent, Codex on ChatGPT Free or GitHub Copilot CLI on Copilot Free. For open source, OpenCode with a free or local model. Darce has a free daily allowance with fast models and no card.

Is Gemini CLI still free?

Not for individual sign-ins. Google stopped serving free Gemini Code Assist for individuals in Gemini CLI on June 18, 2026, and moved individuals to Antigravity CLI, which has a $0 individual plan with basic weekly rate limits. Gemini CLI still works with paid API keys.

Can I run Claude Code with a free local model?

Ollama documents it: run ollama launch claude, or set ANTHROPIC_BASE_URL to http://localhost:11434. Use a model with tool support and at least 64k context. Anthropic doesn't support non-Claude models in Claude Code, so some features may not work.

Related guides