Model comparison
Gemma 4 31B vs Qwen 3.6 27B
Gemma 4 31B vs Qwen 3.6 27B for local inference — VRAM, tokens/sec on RTX 4090 and M4 Max, and quality benchmarks side by side. Qwen 3.6 27B leads on quality, while Qwen 3.6 27B decodes faster on an RTX 4090.
| Metric |  Gemma 4 31B |  Qwen 3.6 27B |
|---|---|---|
| Parameters | 30.7B | 27B |
| Active params (MoE) | dense | dense |
| Context window | 256K | 262K |
| Quality tier | 86 | 99 |
| MMLU-Pro | 85.2 | 86.2 |
| GPQA Diamond | 84.3 | 87.8 |
| SWE-bench Verified | — | 77.2 |
| LiveCodeBench | 80.0 | 83.9 |
| VRAM (Q4_K_M) | 18.7 GB | 16.5 GB |
| Speed — RTX 4090 (tok/s) | 13.0 | 50.4 |
| Speed — M4 Max (tok/s) | 25.5 | 27.4 |
Which should you run?
Qwen 3.6 27B scores higher on the quality benchmarks, so pick it when answer quality matters most. Qwen 3.6 27B is faster to decode on an RTX 4090 (50 vs 13 tok/s). Qwen 3.6 27B needs less VRAM (16.5 GB at Q4_K_M), so it fits on smaller GPUs.