Gemma 4 Broke My AI Agent's Personality (Gemma 4 vs Qwen 3.5)

Gemma 4 vs Qwen 3.5 - I tested Google’s new Gemma 4 31B local LLM against Qwen 3.5 27B as the brain powering one of my OpenClaw AI agents, and the results weren’t what the launch hype promised. In this video I walk through Google’s own published benchmarks, show you exactly what broke, and explain why the benchmark that matters most for local AI agents is one nobody is running.

🖥️ My local AI stack:

• RTX 5090 (32GB VRAM)
• LM Studio for local LLM hosting

Video chapters

• 0:00 — Gemma 4 vs Qwen 3.5
• 0:35 — Benchmarks
• 3:27 — AI Agent Identity
• 5:05 — Comparisons

Watch the full video and find the original resources on YouTube.