Local LLM Benchmarker

One-click benchmark of how fast your Mac runs local LLMs

Problem

Developers want to know real throughput and memory usage of local models on their own machines, but only have scattered personal anecdotes. Someone asked the community for the best local model on Mac with decent speed and posted their own M5 Pro 48GB numbers (Qwen 3.8 Flash at 14-15 tok/s), showing a clear lack of standardized tooling.

Solution

A local benchmark tool that auto-detects hardware, points at user models, runs standard prompts, measures throughput and memory, and generates shareable comparison cards and leaderboards.

Users

Local AI developers, Mac users, model selectors

MVP

Auto-detect Mac chip and memory; support local model paths or common formats; run fixed prompts to measure tok/s and peak memory; generate shareable result card; include built-in model comparison leaderboard

Revenue

Free with paid advanced benchmark templates or history tracking; hardware vendor sponsorship

Why now

Local models are exploding but benchmark tooling is missing; the community already trades anecdotal numbers