Model comparison

Qwen3-Coder-30B-A3B-Instruct vs GLM-4.7-Flash-REAP-23B-A3B-absolute-heresy

At Q4_K_M, GLM-4.7-Flash-REAP-23B-A3B-absolute-heresy is the smaller download — 13,934,220,800 bytes against 18,556,689,568. At long context the gap widens: GLM-4.7-Flash-REAP-23B-A3B-absolute-heresy's KV cache at 32K is 1.8× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

Qwen3-Coder-30B-A3B-InstructGLM-4.7-Flash-REAP-23B-A3B-absolute-heresy
Parameters30.5B23.0B
Architectureqwen3moedeepseek2
Layers4847
Native context262,144202,752
Mixture of expertsyes, 128 expertsyes, 48 experts
Quantizations published4632
Smallest quantization7.46 GiB4.49 GiB
Q4_K_M17.28 GiB12.98 GiB
Licenceapache-2.0mit

KV cache by context

the term that decides long-context viability
ContextQwen3-Coder-30B-A3B-InstructGLM-4.7-Flash-REAP-23B-A3B-absolute-heresyRatio
4,0960.38 GiB0.21 GiB1.82×
8,1920.75 GiB0.41 GiB1.82×
16,3841.50 GiB0.83 GiB1.82×
32,7683.00 GiB1.65 GiB1.82×
65,5366.00 GiB3.30 GiB1.82×
131,07212.00 GiB6.61 GiB1.82×