FitMyLLM

FitMyLLM is an independent benchmark site for self-hosted AI that helps people choose a local model for their hardware. It presents itself as a way to find the best local AI model for a computer, using measured speed, VRAM fit, and quality to rank models against specific GPUs.

The site indexes 328 models and says it benchmarks them across 1,720 GPUs. It shows model recommendations with token-per-second figures, VRAM fit indicators, and a score, and it can return ready-to-run Ollama commands for a chosen model. The homepage also describes use-case filters for chat, coding, reasoning, creative work, vision, roleplay, agentic tasks, and embedding, along with context-length controls and benchmark views for comparing models side by side. A live probe path is shown as /api/recommend, and the interface references a read-only live probe.

Several workflows are described for different audiences. A beginner path detects hardware, picks a use case, and produces a single terminal command. An intermediate path is for users who already know their specs and want to filter and compare models by speed, quality, quantization, and context. A builder path takes a use case and dataset and returns an approach, model choice, training configuration, and deployment commands. An expert path covers multi-GPU sizing, tensor parallelism, total cost of ownership, cloud break-even, and fine-tuning VRAM. The site also offers a “Find My Fit” and comparison flow for models and GPUs.

FitMyLLM is web-based and free to use, with no signup or account required and no paywall. The page also says it is private and has no lock-in. The site includes navigation for Benchmarks, Methodology, GPU Prices, Blog, Guide, and About, and it mentions a weekly briefing.