Claude Code + OpenRouter = Free UNLIMITED Coding (No RAM Needed)
In this video I'll show you how to run Claude Code completely for free using OpenRouter — no local models, no GPU, and no RAM requirements needed. This works on any computer including a Raspberry Pi or VPS. Unlike Ollama setups, everything runs in the cloud so you can access massive 480B parameter models for $0.
🛠️ Hire me: bart@supportlaunchpad.com
👉 Digital Ocean Sign Up: https://m.do.co/c/05f734d27153
--(Get $200 in credit over 60 days for Digital Ocean)
🚀 Sign up to Replit: https://replit.com/refer/bart31
👉 Check out my Skool: https://www.skool.com/flow-state
🗂️ Sign up to Supabase: https://supabase.link/b1VZg8p
🏆 Sign up to Lovable: https://lovable.dev/?via=bart
👉 Sign up to n8n: https://n8n.partnerlinks.io/lyzq2o8cky0o
Also see ComfyUI Tutorial Series_ Ep01 - Introduction and Installation - PLAYLIST
and AI Filmmaking Part 1 _ Video gen with LTX 2.3_ You're doing it wrong.
and LongCat Video Avatar_ Long Talking Avatar Video _ Open Source _ ComfyUI _ Free & Unlimited
Transcript
0:00 · Hello legends. In this video, I'm going to show you how to run Claude Code completely for free. So, inst ead of using one of the paid default models like Sonnet 4.6, we'll be able to plug Claude Code into a free cloud model.
0:11 · This process doesn't need you to install any local models on your device. So, if your computer doesn't have a lot of RAM, or if you've installed Claude Code on your Raspberry Pi or on a VPS, where you literally don't have any space to download any models, then using this process you'll still be able to run Claude Code for free. You'll be able to send in messages and then get a free AI responses. Now, there's already a bunch of really good videos online about how to use Claude Code for free using Ollama and self-hosted LLMs.
0:36 · But, if you take a look at a really popular AI model for coding, even the smallest version of GPT-OSS needs you to have 14 gigs of VRAM on your computer. Now, 14 gigs of VRAM is actually pretty high because let's say you buy a brand new Apple computer, the base RAM that you get is actually 16 gigs. Which means that you still cannot run this model because your computer needs a bunch of RAM in order to run its daily operations. Like if you're browsing the computer or processing video, that still consumes your RAM. Now, if you've watched a bunch of my previous videos, you'd see that I'm using something called OpenRouter for a lot of my AI needs.
1:07 · OpenRouter is a way that you just get one API key and then you can access literally hundreds of different AI models. You can get access to Claude, to GPT, to Gemini, and then so many other models that you probably never heard of. If we go across to the available models page and just type in free into the search bar, we can literally use any of these models with our Claude Code. So, over here we've got the 120 billion parameter version of GPT-OSS, which is completely free with $0 input tokens and $0 output tokens.
1:36 · If we scroll down a little bit further, we also have the Qwen 3 Coder 480 billion parameter model. Once again, completely for free with $0 input and output tokens. And to run something like that Qwen 3 model with the 480 billion parameters, you need to have 290 gigs of VRAM, which is absolutely massive and it would be so expensive that at that stage, you may as well just pay for a Claude Code account. Now, thankfully we can use OpenRouter directly with Claude Code. We can follow some of these instructions.
2:04 · The first instruction just says to install Claude Code on your computer, which we probably already have. Once you've got that installed, we just need to get ourselves an API key.
2:13 · We just need an API key so that we can authenticate the API calls. We're not going to be spending any money at all.
2:19 · And then we just need to follow these instructions and then put these into our settings.json file. Now, there's two ways for us to use OpenRouter with Claude Code. We can either scope it to a local permission, so we create a new folder and we're working within that folder on a project, or we can scope it globally where every time we open up Claude Code in any project, any folder, we'll be using the free OpenRouter. I'm first going to show you how to scope it locally to a project.
2:42 · So, right now I don't have any files on my left-hand side, so I'm just going to create a local .claude folder and inside that folder, I'm going to create a settings.json file. Now, we need to add these four variables into that settings.json file. We need the OpenRouter API key. We need the Anthropic base URL to point across to OpenRouter. We have the Anthropic auth token and then we also need the Anthropic API key to be set to blank.
3:09 · So, this is going to be explicitly empty. And then back in Cursor, paste this into your settings.json file. So, you can either copy this or take a screenshot and drop it into Claude to help you write this out. Let's go get ourselves an API key. If you don't have an account yet, just go ahead and create one with OpenRouter and then go across to API keys and then create an API key.
3:27 · I'm going to call this OpenRouter demo, hit create, copy this key, and then paste it into here. Now, I did read somewhere that you have to have $10 of credit loaded up into your OpenRouter account to use this free method. I've tested it and it works for me when I don't have $10 of credit, but I've been using OpenRouter for a long time and I've and have historically loaded up like hundreds of dollars of credit, so I'm not sure if that will apply to you, but if you're running into any issues, just try that as a troubleshooting step.
3:51 · And now let's get our OpenRouter model.
3:53 · Now, back in the models page, I've typed in free. Let me just choose a model.
3:57 · Let's actually start with the Qwen 3 480 billion parameter model. We'll copy the model name from here and then paste it into here. And now we can use Claude locally with the free Qwen model. So, let's type in Claude and here we've set Qwen 3 Coder as our Claude Code model.
4:11 · Let's type in hi there. Let's type in something. Now, for this specific model, we run into an error and the error is that there's some rate limiting being applied upstream and this just means that there's so many people trying to use this model at this current time, which means that we're being blocked from using it. That is pretty unfortunate, but it is a reality of trying to use some of these free models.
4:29 · So, I'm just going to exit out of here and choose a different model. Now, I've used this process a few times and it is a little bit hit and miss with whichever model that you use, but if you scroll to the very, very bottom of this free models page, you actually see this general free models router. If we click into here, so this is actually a request router. As we hit that endpoint, OpenRouter will take our request and then route it to one of these models. For example, we can see that there's two Nvidia models part of this batch.
4:53 · But, if we just click this very top model over here, Step 3.5 Flash, and just check out some of the specs, it looks like for artificial analysis, it's in the top 20% of models compared. For coding, it's also in the top 20% and then for a GenTech index, it's also actually in the top like 10%.
5:11 · And then scrolling down a little bit more, you can see the main use cases for this model, which number one is OpenClaude with 14.5 billion tokens consumed. And then over here we have number four with Claude Code. So, it looks like it's a pretty popular model for OpenClaude and Claude Code. So, let's just take this model, copy it, and let's replace the Qwen model with this new one. Let's go back into our terminal and relaunch Claude Code. And we can see over here we've got the OpenRouter-free.
5:37 · Let's test it out.
5:39 · Hi there. Awesome. And here's our response. Now, using the free model router, you're going to have a lot more success and pretty much all your requests will be going through. Now, the final thing for us to do is figure out how to get the settings applied to every project we work on. So, I've just opened up a terminal on my computer. I've just launched Claude Code and we can see here we're using the Sonnet 4.6. So, now we have to find the global Claude configuration file. So, over here we created something locally for this specific folder in this project. I'm just going to run this command to open up my global Claude folder and here we go, we've just opened up that folder.
6:11 · Now, we want to take this settings.json file and open up in Cursor. So, I'm just going to drag that file and drop it into Cursor. And then here we can see the global configuration that I've set up.
6:20 · All you need to do is go back to your local configuration file, copy this, and let's drop it into the environment variables block. So, I'm just going to put a comma here cuz I'm going to be setting some more variables, paste in our variables from before, and now we can use the free version for every Claude Code session. Going back to our terminal, let's just exit out of this session.
6:39 · Let's re-enter into a new one. And here we go, we've got OpenRouter/free as our main model. Let's send in a message. And here is that response. Now, I just want to say that using these local models, the performance is not going to be anywhere near as good as using something like Sonnet 4.6 or Opus 4.6. Even if you manage to install something like the 20 billion parameter model or the 120 billion parameter model, these are going to be good locally, but very demanding on your machine, but they're still not going to be able to keep up with Sonnet 4.6 or Opus 4.6.
7:04 · Now, since we're using MiniMax, we can literally flip out to any model that we want, even a paid model. So, we've got over here GPT 5.4 Pro. We can literally use this in Claude Code if we wanted to, but a really popular model to use for AI coding is the MiniMax 2.5. If we click into this and just drop down on these performance categories, you can see that it's rated number one for programming. And if you want to use the MiniMax model, you can just come and copy this model name and just like we've been doing, just replace it here, spool up another Claude Code session, and we've got MiniMax M2.5.
7:39 · Let's test it out. Hi there.
7:41 · And we've got our response. It's literally that easy to flip between models. Now, the reason MiniMax is so popular is that because in some recent benchmarking, the MiniMax, which is this red column over here, compared to something like the Opus 4.6, which is the lighter gray, which is something like this, it performed extremely well compared to the Opus 4.6 model for coding tasks. And if you look at the costs, it's 30 cents per million input tokens, $1.20 per million output versus Opus 4.6, which is $5 per million input tokens and $25 per million output tokens. So, it's so much cheaper and will give you pretty comparable results.
8:13 · All right, thank you very much for watching and I'll see you in the next one.