Benchmarks · community-measured

Apple Silicon LLM benchmarks

Name: Rapid-MLX community benchmarks
Creator: Rapid-MLX
License: https://www.apache.org/licenses/LICENSE-2.0

Measured local-inference performance, contributed by rapid-mlx users: 35 model-and-chip combinations from 58 submissions across 11 machines (updated 2026-09-27). Explore and filter on the interactive leaderboard, or jump straight to a measurement below. Every number is reproducible: rapid-mlx bench <model> --tier speed --submit.

Interactive leaderboard → Raw corpus /api/benchmarks

Updated 2026-09-27

model-and-chip
combinations

submissions

machines

Submissions are anonymous and retractable — see the leaderboard for details.

One command, any Mac

Every number here is reproducible

$ rapid-mlx bench <model> --tier speed --submit

Per page

bonsai-1.7b-2bit2 chips measured 329tok/sApple M5 Max Apple M4 Max 315 bonsai-27b-2bit1 chip measured 3.8tok/sApple M2 gemma-4-26b-4bit2 chips measured 108tok/sApple M4 Max Apple M4 Pro 72.3 gemma-4-26b-qat-4bit1 chip measured 66.7tok/sApple M4 Pro gpt-oss-20b-mxfp4-q82 chips measured 107tok/sApple M5 Max Apple M2 Pro 44.6 lfm2.5-1b-4bit1 chip measured 219tok/sApple M1 Max qwen2.5-14b-4bit2 chips measured 45.8tok/sApple M4 Max Apple M5 Max 43.5 qwen3-8b-4bit1 chip measured 48.3tok/sApple M4 Pro qwen3.5-27b-4bit1 chip measured 25.3tok/sApple M5 Max qwen3.5-4b-4bit7 chips measured 138tok/sApple M4 Max Apple M5 Max 130Apple M1 Max 90.4Apple M4 Pro 82.8Apple M2 Pro 56.4Apple M4 37.3Apple M2 15.5 qwen3.5-9b-4bit3 chips measured 78.9tok/sApple M5 Max Apple M4 Pro 49.2Apple M2 Pro 34.6 qwen3.5-9b-8bit1 chip measured 45.2tok/sApple M5 Max qwen3.6-27b-4bit1 chip measured 25.9tok/sApple M4 Max qwen3.6-35b-4bit3 chips measured 91.9tok/sApple M3 Ultra Apple M5 Pro 86.3Apple M4 Pro 85.5 qwen3.6-35b-optiq-4bit2 chips measured 92.9tok/sApple M5 Max Apple M4 Pro 64.4 qwen3.8-27b-4bit3 chips measured 28.9tok/sApple M4 Max · mtp Apple M5 Max 25.3Apple M4 Pro 15.2 qwen3.8-27b-mixed-3.5bpw2 chips measured 26.5tok/sApple M5 Max Apple M4 Pro 15.9

No model or chip matches

35 model-and-chip combinations exist in this corpus. Clear the search to bring every group back.

Two commands

Add your machine

Install the engine, run the standardized speed tier, and your Mac joins the corpus above.

Submissions are anonymous and retractable — see the leaderboard for details.

1 · Install the engine

$ curl -fsSL https://rapidmlx.com/install.sh | bash

2 · Run the speed tier and submit

$ rapid-mlx bench qwen3.5-9b-4bit --tier speed --submit

Raw corpus: /api/benchmarks · Submissions are anonymous and retractable — see the leaderboard for details.