Submit a result
Share a benchmark you ran (or found). Every accepted result is credited on the contributor leaderboard. A public link to the raw output (llama-bench table, gist, forum post) is required.
How to get raw output (one command)
If you use llama.cpp, run this in the folder with your model and paste everything it prints:
llama-bench -m model.gguf -p 512 -n 128 -t 4-m is your model file · -p 512 measures reading a 512-token prompt · -n 128 measures generating 128 tokens (the "Gen t/s" we rank by) · -t 4 uses 4 CPU threads (Pi 5 has 4 cores). Using Ollama instead? Run ollama run MODEL --verbose, ask a question, and paste the stats printed at the end.
Results without pasted raw output are not added. We compare the numbers you type against the output before publishing.