I Tested a 100M AI Model on a 24GB RAM Laptop
About This Video
How small can an AI model actually be? In this video, I test a **100 million parameter AI model** on my **24GB RAM laptop** to see how well it performs, how much memory it uses, how fast it responds, and—most importantly—whether a model this small is actually useful. This is Episode 1 of a series where I test AI models from the smallest to the largest. We're starting at just **100M parameters** and eventually working our way up to much larger models to find out exactly how far a 24GB RAM laptop can go. In this episode, I test: • 100M parameter AI model • Local AI performance • RAM usage • Response speed • Real-world prompts • Model quality • Whether it's actually useful • How a tiny model compares to larger AI models The goal isn't just to see whether the model runs. **The goal is to find out what you actually get when you go this small.** And this is only the beginning. 100M → 300M → 500M → 1B → 3B → 7B → 14B → and beyond. Subscribe if you want to follow the entire experiment.
Related Videos
OmniRoute Setup on Mac: Connect Claude Code for Free AI Tokens
earn how to install and configure OmniRoute on macOS to route Claude Code through free AI model endpoints and avoid token rate limits. Complete Terminal guide. <p class="lead">OmniRoute is an open-source local AI gateway that lets you aggregate free-tier provider models behind a single OpenAI-compatible endpoint. Follow this step-by-step guide to install OmniRoute on macOS and integrate it with Claude Code to bypass usage limits.</p> <h2>Prerequisites</h2> <ul> <li>A Mac running macOS Big Sur or newer (Intel or Apple Silicon M1/M2/M3/M4)</li> <li>Node.js (v18 or higher recommended)</li> <li>Terminal app (or iTerm2)</li> </ul> <h2>Step 1: Install OmniRoute Globally on Mac</h2> <p>Open your macOS Terminal and execute the global npm installation command:</p> <code>npm i -g omniroute</code> <h2>Step 2: Start the OmniRoute Service</h2> <p>Launch the OmniRoute gateway server locally:</p> <code>omniroute</code> <p>This starts the gateway on your local machine (defaulting to port 20128) and prints the local dashboard URL: http://localhost:20128.</p> <h2>Step 3: Connect Free Providers in the Dashboard</h2> <ol> <li>Open http://localhost:20128 in your browser.</li> <li>Navigate to the Providers tab.</li> <li>Connect your preferred free-tier API providers (such as Gemini, Groq, Kimi, GLM, or DeepSeek).</li> <li>Go to the API Keys section and generate a local token for authentication.</li> </ol> <h2>Step 4: Configure Claude Code to Route Through OmniRoute</h2> <p>To direct Claude Code CLI traffic through your local OmniRoute instance, set the Anthropic environment variables in your active shell or ~/.zshrc file:</p> <code>export ANTHROPIC_BASE_URL="http://localhost:20128"</code><br> <code>export ANTHROPIC_AUTH_TOKEN="your_omniroute_api_key"</code><br> <code>export CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1</code> <h2>Step 5: Verify the Connection</h2> <p>Run a test query in Claude Code to verify requests are reaching your gateway:</p> <code>claude "Hello, confirm connection status."</code> <h2>Useful Resources</h2> <ul> <li><a href="https://github.com/diegosouzapw/OmniRoute" target="_blank" rel="noopener">OmniRoute Official GitHub Repository</a></li> <li><a href="https://nodejs.org/" target="_blank" rel="noopener">Download Node.js for Mac</a></li> <li><a href="https://docs.anthropic.com/en/docs/agents-and-tools/claude-code/overview" target="_blank" rel="noopener">Claude Code Official Documentation</a></li> </ul>
Use GPT 6 Astra FREE in Codex (Free Unlimited Codex with GPT 6)
In this video, I'm walking you through how to set up and configure the GPT-6 Astra model with your local development environment using Flagship Router. This is an educational tutorial about local model configuration and development tool management—all standard practices that developers use for their workflows. 📌 What This Video Covers: Cloning the Flagship Router repository from the official source Installing Node.js and project dependencies with npm install Running the launch command to build and start the local router Logging into the Flagship Router web interface Turning off the API key requirement for smoother operation Exploring the model library—336 models available with 86 ready to use Finding and testing the GPT-6 Astra model Copying the model identifier for configuration Using the CLI Tools section to configure your development application Applying the model configuration and launching your app Verifying the setup with a live test prompt Confirming the connection through Flagship Router ⚙️ How It Works: Flagship Router is a local routing tool that helps developers manage connections between their coding applications and various model providers. In this tutorial, I demonstrate how to configure it with the GPT-6 Astra model and integrate it with development tools using standard local proxy techniques. 🛠️ Prerequisites: Node.js installed on your system Git installed for cloning the repository Basic familiarity with terminal or command prompt A development application or code editor installed 🔗 Resources Mentioned: Flagship Router Repository: https://github.com/theRizwan/FlagshipRouter Node.js Official Website: https://nodejs.org 📝 Disclaimer: This video is for educational and informational purposes only. It demonstrates standard local development environment configuration techniques using open-source tools and models. The tools and methods shown are intended for personal learning and legitimate development workflows. Always ensure you comply with the terms of service of any software or platform you use. 💬 Have questions about the setup or ran into an issue? Drop a comment below and I'll personally help you out. 🔔 Subscribe to Rizwan for more coding tutorials, development environment setups, and tech guides every week. #GPT6Astra #FlagshipRouter #LocalDevelopment #CodingSetup #AIModels #ProgrammingTutorial #Rizwan
I Tested 1B-1.7B AI Models Locally — Here Is What Happened
I ran Claude on 1B–1.7B local AI models to see if they could actually handle coding tasks on my 24GB RAM machine. Testing small language models is the best way to understand the limits of local hardware. Monitoring system performance while pushing these tiny models to their breaking point reveals if they can maintain efficiency without killing CPU or memory resources. Whether you are using a Claude coding agent or exploring general LLM inference, knowing which model size fits your specific setup is key. The results from these tests highlight which configurations work and where the performance falls off. Subscribe for weekly AI performance breakdowns, and let me know in the comments if you want to see a test with larger parameter sizes next.