VDAOvideo

I tested 3 local AI models. The smallest one won.

Joyce Lin · 53.1K views · 6 months ago
Description

The #1 AI models are all locked behind APIs. So I tested the best ones you can actually run locally — on a Mac Mini, no cloud, no data leaving your machine. I compared Llama, Qwen, and Gemma across three tests: logic, technical knowledge, and a real-world question. The smallest model won. Here's how to pick the right local AI model for your hardware and use case. 00:00 The problem with top AI models 00:23 Filter #1: Open weights only 01:01 Filter #2: What size model can you run? 01:39 Filter #3: Quantization & quality tradeoffs 02:08 The models I picked (Llama, Qwen, Gemma) 02:54 Setup with Ollama 03:18 Test 1: Logic puzzle 04:20 Test 2: Technical explanation 04:57 Test 3: Real-world question 06:09 Results: Speed 06:30 Results: Accuracy 06:43 Results: Effort to get a useful answer 07:04