A new web page has been developed to help users determine which AI models their local hardware can run. The tool analyzes a laptop's configuration, including RAM, to estimate its AI hosting capability across different model tiers: Entry (1B), Mid (7B), Large (70B), and SOTA (1T+). It provides a comparison of model tiers with example models and their estimated RAM requirements.

This tool aims to inform users about their local AI potential, showing how far they are from running top-tier SOTA models. The estimates are based on quantized models (4-bit) suitable for consumer hardware. Actual performance can vary based on the quality of quantization, the specific model architecture, and the optimization tools used, such as llama.cpp or vLLM.

While SOTA models like GPT-4o, Claude 3.5, and Gemini Ultra are estimated to be around 1T+ parameters, this checker helps users understand what's feasible locally today. It offers a practical way to gauge the capabilities of personal devices for running AI models without relying solely on cloud-based solutions.