Description
LocalScore is an open benchmark that measures how well your computer runs local AI models. You run the benchmark on your own hardware and it reports a LOCALSCORE along with three measures: prompt processing speed in tokens per second, text generation speed in tokens per second, and time to first token in milliseconds. Results are grouped by model size, including tiny (about 1 billion parameters), small (about 8 billion) and medium (about 14 billion), and by whether they ran on a GPU, CPU and GPU, or CPU only. The site has a leaderboard of the latest submitted results, with entries for graphics cards such as the RTX 4090, H100 and A100 and for laptop GPUs. The project is open source, with its code on GitHub as part of the llamafile project. The site also has a download page and a Discord community for discussion.…
Highlights
- Benchmark for local AI on your own computer
- Measures prompt speed, generation speed and first-token time
- Results by model size and by GPU or CPU
- Public leaderboard of latest results
- Open source on GitHub
Ratings & Reviews
No ratings yet
Be the first to rate LocalScore.
People also visit
LocalScore FAQ
What is LocalScore?
An open benchmark that helps you understand how well your computer can handle local AI tasks.
What does a LOCALSCORE mean?
The site combines prompt speed, generation speed and time to first token into one score, so higher numbers mean the hardware runs the model faster. The site shows the individual figures alongside it.
Which model sizes are tested?
The leaderboard groups results into tiny (about 1 billion parameters), small (about 8 billion) and medium (about 14 billion) model sizes.
Where can I see results?
The home page shows a leaderboard of results, and a Latest Results page lists recent submissions.
Is it open source?
Yes. The benchmark's code is on GitHub as part of the llamafile project, and the site links to the repository and its issues.
What are the best LocalScore alternatives?
Popular alternatives to LocalScore include Anubis Oss, Artificial Analysis and whatcani.run. See all LocalScore alternatives




