Show HN: Find the best local LLM for your hardware, ranked by benchmarks

A new tool called “whichllm” has emerged, designed to help users identify the most suitable local large language models (LLMs) for their specific hardware configurations. By auto-detecting GPU, CPU, and RAM, whichllm ranks models from HuggingFace based on performance benchmarks, making it easier for users to select the best option for their needs. This development is timely as the demand for efficient and tailored AI solutions continues to grow, especially among developers and businesses looking to optimize their AI deployments.

The tool features an evidence-based ranking system that prioritizes models not just by size but by real-world performance metrics. For example, the Qwen/Qwen3.6-27B model ranks first for an RTX 4090 GPU with a score of 92.8 and a speed of 27 tokens per second, despite being smaller than other contenders. This nuanced approach addresses a common challenge in the AI community: finding the best-performing model that fits within specific hardware limitations. Users can simulate different GPUs to see how various models perform, ensuring they make informed decisions before investing in hardware or software.

What sets whichllm apart is its commitment to accuracy and recency, with live data pulled directly from HuggingFace and a transparent scoring system that discounts potentially inflated claims. This focus on real-time performance and evidence-based recommendations could shift how developers approach model selection, fostering a more data-driven environment in the AI landscape.

As the tool gains traction, it will be interesting to see how it influences the competitive landscape of LLMs and whether it encourages further innovation in model efficiency and accessibility.