Cost & tokens

Keep Claude Code Running on Free AI Models When Your Limit Hits

7 minute readUpdated September 2026Explore more

TL;DR

You can now run Claude Code completely free with almost unlimited usage, as long as you know what that means. Claude Code is the app, and the brain behind it can be swapped for free models like Gemini, GLM and Kimi. A free GitHub list tracks 16 companies with free AI models, and Manifest, a free tool from the same team, stacks them so Claude Code jumps to the next model when one hits its limit. These models are not Claude, and every free plan still has caps, so use them for small jobs and keep your real builds on Claude.

Learn Claude Code. Earn income. Only $9.

👉 https://www.skool.com/claudecodeclub

What running Claude Code for free really means

Claude Code is the app you type into. The model is the brain behind it. Normally that brain is Claude, and your plan's usage limit decides how long you can work.

Claude Code can also talk to a gateway, a middleman that sends your requests to other models. Point it at a gateway full of free models and Claude Code keeps working at no cost. The catch: those models are Gemini, GLM, Kimi and friends, not Claude. Claude's own models are never free here.

Step 1: Get free keys from the Awesome Free LLM APIs list

The Awesome Free LLM APIs repo on GitHub with its list of free AI providers
Awesome Free LLM APIs by the Manifest team on GitHub

Awesome Free LLM APIs is a list of 16 companies that give AI models away with a permanent free plan and no credit card. Each entry links straight to the page where you make an API key, which is a password that lets a tool use that company's models.

Where to start: grab 2 or 3 keys, not 16. Google Gemini (Google AI Studio), Z AI for GLM-4.7-Flash, and Groq are easy sign-ups.

repo URLhttps://github.com/mnfst/awesome-free-llm-apis
promptRead this list: https://github.com/mnfst/awesome-free-llm-apis
I want free models to use with Claude Code for small coding jobs. Pick the 3 best providers for me, tell me why, and give me the exact page to get each API key. Skip anything that is non-commercial only.

Step 2: Set up Manifest to stack your free models

The Manifest LLM Gateway repo on GitHub
Manifest LLM Gateway on GitHub (MIT license)

Manifest is a free, open source gateway from the same team. You add your free keys once, and it gives you one address and one Manifest key. When a model fails or hits its limit, Manifest tries the next one on your list.

How to install Manifest: copy the repo URL below, paste it into Claude Code, and say: install this. It runs in Docker, a free app that runs tools in their own little box. No install at all? The cloud version at app.manifest.build has a free Starter plan.

repo URLhttps://github.com/mnfst/llm-gateway

Or install it by hand (from the README), then open http://localhost:2099 and sign up. The first account becomes the admin.

bashbash <(curl -sSL https://raw.githubusercontent.com/mnfst/llm-gateway/main/docker/install.sh)

In the Manifest dashboard, add the free keys from Step 1 as providers, then pick Claude Code as your agent. Manifest shows you a small settings block with your address and your Manifest key.

Step 3: Make a free mode for Claude Code

Manifest's setup tells you to paste its block into your main Claude Code settings. Don't. That switches every session off Claude for good. Put it in its own file instead, so free mode is something you turn on only when your limit hits.

promptCreate a file at ~/.claude/free-mode.json with exactly this, using my Manifest address and key:

{
  "model": "auto",
  "env": {
    "ANTHROPIC_BASE_URL": "[your Manifest address]",
    "ANTHROPIC_AUTH_TOKEN": "[your Manifest key]"
  }
}

Do not change my main settings.json. Then tell me the one command to start Claude Code in free mode.

When your Claude limit hits, start a new session in free mode:

bashclaude --settings ~/.claude/free-mode.json

Your normal `claude` command stays on Claude, same as before.

The honest catch

  • These are not Claude. Anthropic says it doesn't support running Claude Code on non-Claude models, so some features can act up. Use free mode for small jobs: quick fixes, renames, questions about your files.
  • Free is not unlimited. Every free plan has caps. On the list today, Gemini 2.5 Pro allows 50 requests a day, OpenRouter's free models 50 a day, and GLM-4.7-Flash one request at a time. Stacking helps, but a long session can still run dry.
  • Some free plans train on your chats. Google's free tier, Mistral's free mode and some OpenRouter providers may use your prompts. Keep client files and passwords out of free mode.
  • Never connect your Claude Pro or Max login to Manifest. Manifest offers it, but Anthropic's rules don't allow sending your subscription through a third-party tool. Your account is the thing at risk.

Which setup to start with

  • Want zero installs: Manifest cloud (free Starter plan) plus one Google Gemini key.
  • Want it all on your computer: Manifest in Docker plus Gemini, GLM and Groq keys.
  • Hit your limit once a month: skip all of this and wait for the reset. Free mode is for people who hit it every week.
Learn Claude Code. Earn income. Only $9.

👉 https://www.skool.com/claudecodeclub

Common questions

  • Is Claude free with this setup?

    No. Claude Code, the app, runs for free, but the answers come from free models like Gemini, GLM and Kimi. Claude's own models still need your plan or an Anthropic API key.

  • Is this against Anthropic's rules?

    Pointing Claude Code at a gateway with your own keys is a normal, documented setting. Anthropic doesn't support non-Claude models there, so features may break. What Anthropic does not allow is routing your Free, Pro or Max login through a third-party tool, so never connect your Claude subscription to Manifest.

  • Does Manifest switch over the moment my Claude limit runs out?

    Only if Claude is inside Manifest, and the only safe way to do that is a paid Anthropic API key. With the free setup here, you switch yourself with one command, and Manifest handles the jumps between free models.

  • Is it really unlimited?

    No. Every free plan has daily or per-minute caps. Stacking several providers gets you a lot more room, not infinite room.

  • How do I go back to normal Claude Code?

    Just run claude as usual. Free mode only turns on when you start with the --settings free-mode.json command.

Want the skills that make Claude Code worth every token?

Get 650+ plug-and-play skills, MCPs & prompts, plus 8,000+ members - $9/mo, cancel anytime.

Join the Club