Model comparison
gemma-4-E4B-Agentic-Opus-Reasoning-GeminiCLI-mlx-4bit vs Qwen3.8-27B
At Q4_K_M, gemma-4-E4B-Agentic-Opus-Reasoning-GeminiCLI-mlx-4bit is the smaller download — 5,335,285,280 bytes against 16,810,714,336. At long context the gap widens: gemma-4-E4B-Agentic-Opus-Reasoning-GeminiCLI-mlx-4bit's KV cache at 32K is 4.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| gemma-4-E4B-Agentic-Opus-Reasoning-GeminiCLI-mlx-4bit | Qwen3.8-27B | |
|---|---|---|
| Parameters | 7.5B | 27.8B |
| Architecture | gemma4 | qwen35 |
| Layers | 42 | 64 |
| Native context | 131,072 | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 4 | 22 |
| Smallest quantization | 4.97 GiB | 8.39 GiB |
| Q4_K_M | 4.97 GiB | 15.66 GiB |
| Licence | gemma | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | gemma-4-E4B-Agentic-Opus-Reasoning-GeminiCLI-mlx-4bit | Qwen3.8-27B | Ratio |
|---|---|---|---|
| 4,096 | 0.12 GiB | 0.25 GiB | 2.03× |
| 8,192 | 0.18 GiB | 0.50 GiB | 2.81× |
| 16,384 | 0.29 GiB | 1.00 GiB | 3.48× |
| 32,768 | 0.51 GiB | 2.00 GiB | 3.95× |
| 65,536 | 0.94 GiB | 4.00 GiB | 4.24× |
| 131,072 | 1.82 GiB | 8.00 GiB | 4.40× |