Is Gemma 4 the Qwen Killer? (Tested on a Pi 5)

Gemma 4 vs Qwen 3.5 on a Raspberry Pi 5 — I ran both 2B models head-to-head on the same $100 board. Text benchmarks, image recognition, and the new thinking mode. One of them is almost twice as fast as the other.

In this video I test Google's brand new Gemma 4 E2B and Alibaba's Qwen 3.5 2B running locally on a Raspberry Pi 5 with an 8TB NVMe SSD. Both models are 2-billion-parameter class, both run on Ollama, and both claim "frontier 2B" performance — but only one of them actually delivers on the Pi. Full benchmark suite: raw tokens/sec, prompt quality, image classification on real photos, and thinking-mode vs non-thinking-mode comparison.

🔗 LINKS
- Gemma 4 E2B on Ollama: https://ollama.com/library/gemma4:e2b
- Qwen 3.5 2B on Ollama: https://ollama.com/library/qwen3.5:2b
⏱ CHAPTERS
0:00 Intro
1:12 Installing Gemma 4 on Pi and first test
4:32 Gemma 4 token/s
5: 09 Comparison with Qwen 3.5
7:09 benchmark test between Gemma 4 and Qwen 3.5
12:04 Outro
🏷 TAGS
raspberry pi 5, gemma 4, qwen 3.5, local ai, ollama raspberry pi, gemma 4 e2b, qwen 3.5 2b, raspberry pi llm, local llm pi 5, gemma 4 vs qwen, raspberry pi ai server, nvme raspberry pi

#RaspberryPi #LocalAI #Gemma4

Видео Is Gemma 4 the Qwen Killer? (Tested on a Pi 5) канала Mayukh Builds