Local LLM Finder

Share a test

Hardware, model and outcome are enough. No account needed.

Import speed or benchmark stats

Paste a speed line or choose a stats-only file. Raw text stays in your browser.

Need a test prompt? optional

Use a fresh chat in your usual app. No model response needs to be submitted.

Compare repeat runs
  1. Keep the model, quant and GPU/CPU settings the same. If available, use a 4096-token context, 128-token reply limit, temperature 0 and consistent thinking mode.
  2. Run once to warm up, then three times in fresh chats. Report the middle generation speed; skip it if your app doesn’t show one.

Manual reports are observations. Only reviewed, matching repeated measurements can inform calibration.

Run an Ollama benchmark instead

The Python helper runs a warm-up and five tests with seed 42. It saves stats locally, without responses or automatic uploads.

Download helper

Windows: py llf-test.py. Mac/Linux: python3 llf-test.py. Add --prompt code for coding. Import the saved JSON using “Import speed or benchmark stats”.

How did it run?
tokens / second

Reply speed, not prompt processing. Leave blank if unknown.

Add settings or a note optional

Submitting allows publication and use in estimates after review. Keep private information out of notes.