Add your machine's real numbers

Most of our speeds are estimates. Every real run replaces one with a measured number, for everyone. It takes about two minutes if you have Ollama installed.

Run this

curl -fsSL https://vsmacs.com/bench.sh | sh

Want a different model? curl -fsSL https://vsmacs.com/bench.sh | sh -s -- llama3.1:8b. Read the script first: it's 25 lines.

What it sends

The model name, Ollama's own timing numbers (tokens/sec, load time), your chip and GPU name, total RAM and OS version. No IP-derived location is stored, no files, no hostname. We review every submission by hand before it appears on the site.

No Ollama?

Email your machine, the model and the tokens/sec your tool reports (LM Studio, llama.cpp, MLX all work) to [email protected].