Your computer
has a type.
Find local AI models that fit your computer’s memory, runtime, and context.
Automatic checks for new and trending models on every visit, and every 15 minutes while open. Major releases enter through the same live feeds as new fine-tunes and quants.
How updates work
Available MLX and CUDA variants appear as their public metadata is discovered. New releases may take time to gain compatible variants or benchmark scores. Benchmark sources refresh every 6 hours; unavailable sources keep their last dated results.
Confirm your computer’s memory before relying on these results.
Recommended for you
Example matches. Confirm your setup to make these yours.
Hot picks
Live Hugging Face trends for your compute platform.
Memory decides what fits. Intelligence is a published test score. Speed is an estimated output rate. Open any model for Hugging Face links, install instructions, and the evidence.
25 sourced benchmark results · Benchmark sources refresh every 6 hours on use.
Local model fit, explained
How does LocalModelMatch estimate memory?
It compares each model variant’s estimated weights, context cache, and runtime allowance with the memory budget you set. Browser hardware checks cannot confirm exact installed memory, so verify your device settings before relying on a fit.
Does a comfortable fit guarantee good speed?
No. Fit is a memory estimate. Published MMLU-Pro scores are references, and output-speed ranges are bandwidth estimates; neither measures your computer running that model variant.
Does LocalModelMatch run models or upload hardware data?
No. The site does not execute models or host model weights. Setup choices stay in this browser. The catalog service receives only the model query and runtime; hardware detection is best-effort.
Can I compare a computer before I buy it?
Yes. Use What If to compare sample computers, memory budgets, and context sizes before you choose hardware. These estimates do not guarantee performance or runtime compatibility. Open the What If planner.