Gemma 4 Edges Qwen3.5 on SVGs

This is a noisy but useful early signal: Gemma 4 seems to be hitting above its weight on agentic and code-adjacent tasks, not just prose.

• –The test used Q4 quants from Unsloth, so treat it as a practical smoke test rather than a controlled benchmark
• –The result matches Google’s positioning for Gemma 4: native function calling, structured output, code generation, and multilingual strength
• –Qwen3.5 still appears competitive on aesthetic output, with commenters preferring its style even when Gemma captured details better
• –For SVG generation, fidelity and editability may matter more than raw creativity, which favors the model that preserves composition under iterative prompting
• –A better follow-up would be image-feedback iterations, since one-pass SVG prompts can hide weaknesses in revision and self-correction