MacBook Neo Local AI Test – LLM Benchmarks & MLX Performance!

Timestamps:

00:00 – Intro
01:47 – Qwen3.5 4B Test
05:24 – LiquidAI LFM1.2 Test
07:00 – Qwen3.5 9B Test
08:03 – Qwen3.5 0.8B Test
09:52 – Llama3 8B Test
11:35 – Claude Code System Optimization
12:57 – Vibe Optimizing With Claude
14:17 – Letting Claude Run Benchmarks
15:21 – MLX Llama3 8B Speed Testing
18:15 – Speculative Decoding Test
19:05 – MLX Llama3 8B Benchmark Results
20:55 – Claude Benchmark Script Demo
22:37 – Closing Thoughts

AI Integration & Consulting: https://bijanbowen.com/
Join the Discord: https://discord.gg/hfaR2exy7S

In this video, we test how well the MacBook Neo performs for local AI workloads. Several models are benchmarked, including Qwen3.5 (0.8B, 4B, and 9B), LiquidAI LFM1.2, and Llama 3 8B, with a focus on real inference performance.

We also experiment with Claude Code to optimize the system and automatically generate benchmark scripts. The tests include MLX-based inference speed measurements and speculative decoding experiments to see how much performance can be improved through optimization.

source