Surprising Performance of Small Qwen3-A3B MoE - Testing on Extreme Logic Test

This video demonstrates the performance testing of the new MoE Qwen3-30B-A3B model on an extreme logic test, comparing results with "thinking mode" both active and deactivated. Explore how a Mixture of Experts (MoE) model with only 3B active parameters handles highly complex logical reasoning challenges in this live recording from Discover AI. The 32-minute demonstration includes testing both local Ollama implementation and online versions of the model, with interesting results when the model is corrected and when thinking mode is activated. For comparison, a link to testing of the larger Qwen3-235B-A22B model on the same test is provided. The video is organized into chapters covering the test introduction, local and online implementations, error correction, thinking mode activation, and final verification.