Before you download another 40GB model, check your fit

What it does

llmfit is a Rust-based terminal utility that inspects your machine’s RAM, CPU, and GPU, then compares those specs against a catalog of hundreds of local models. It returns a composite score for each model across quality, speed, memory fit, and context length, and suggests the best quantization level and run mode. The default interface is an interactive TUI, though a classic CLI mode is available for scripting.

The interesting bit

The tool flips the usual “download and pray” workflow. A Plan mode inverts the problem: pick a model and it tells you what hardware you would need to hit a target token-per-second speed. More unusually, it aggregates real-world performance measurements from a community leaderboard spanning 27+ hardware presets, so you can compare estimates against actual user benchmarks before you buy or build.

Key highlights

• Auto-detects multi-GPU, MoE architectures, and mixed CPU/GPU setups
• Ranks models by fit categories: Perfect, Good, Marginal, or Runnable
• Vim-inspired TUI with Normal, Visual, and Select modes for bulk comparison and column-based filtering
• Integrates with local runtimes including Ollama, llama.cpp, MLX, LM Studio, and Docker Model Runner
• Community leaderboard powered by localmaxxing.com with measured tok/s, TTFT, and VRAM across presets from Apple M1 to RTX 5090

Verdict

Worth a look if you maintain a local model zoo or are sizing a new rig. Probably overkill if you only ever run one small model via a single chat app.