A hands-on comparison of Qwen 3.6 (35B MoE) vs Gemma 4 (31B) running locally on an Apple M5 Max with 128GB RAM using LM Studio. Tests include SVG image generation and a browser-based snake game. Qwen 3.6 consistently outperforms Gemma 4 in inference speed (~110 tokens/sec vs ~16 tokens/sec), thinking time (under 1 second vs 25-105 seconds), and output quality. Both models are vision-capable, but Qwen 3.6 produced more complete and functional code outputs.

15m watch time
11 Impressions