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-4bitQwen3.8-27B
Parameters7.5B27.8B
Architecturegemma4qwen35
Layers4264
Native context131,072262,144
Mixture of expertsnono
Quantizations published422
Smallest quantization4.97 GiB8.39 GiB
Q4_K_M4.97 GiB15.66 GiB
Licencegemmaapache-2.0

KV cache by context

the term that decides long-context viability
Contextgemma-4-E4B-Agentic-Opus-Reasoning-GeminiCLI-mlx-4bitQwen3.8-27BRatio
4,0960.12 GiB0.25 GiB2.03×
8,1920.18 GiB0.50 GiB2.81×
16,3840.29 GiB1.00 GiB3.48×
32,7680.51 GiB2.00 GiB3.95×
65,5360.94 GiB4.00 GiB4.24×
131,0721.82 GiB8.00 GiB4.40×