Qwen3.6-35B vs Gemma4-26B on 7900 XTX

In three linesBenchmark on Radeon 7900 XTX: Qwen3.6-35B vs Gemma4-26B with reasoning enabled. Qwen generates 2x more tokens (14,811 vs 7,386) but Gemma is ~20% faster end-to-end (95.6s vs 118.8s). Qwen's MTP reaches 130 tok/s vs 78 tok/s, but token count becomes the bottleneck. Quality close, interesting per-task splits.

Summary generated by Claude — human-verified