Is Mellum2-12B-A2.5B-Thinking or Qwen3-Coder-30B-A3B-Instruct smaller?

Mellum2-12B-A2.5B-Thinking is smaller at Q4_K_M: 7.52 GiB against 17.28 GiB.

Which uses less memory at long context?

Mellum2-12B-A2.5B-Thinking — its KV cache at 32K context is 6.0x smaller, because of how its attention layers cache.

Model comparison

Mellum2-12B-A2.5B-Thinking vs Qwen3-Coder-30B-A3B-Instruct

At Q4_K_M, Mellum2-12B-A2.5B-Thinking is the smaller download — 8,071,295,040 bytes against 18,556,689,568. At long context the gap widens: Mellum2-12B-A2.5B-Thinking's KV cache at 32K is 6.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Mellum2-12B-A2.5B-Thinking	Qwen3-Coder-30B-A3B-Instruct
Parameters	12.1B	30.5B
Architecture	mellum	qwen3moe
Layers	28	48
Native context	131,072	262,144
Mixture of experts	yes, 64 experts	yes, 128 experts
Quantizations published	3	46
Smallest quantization	7.52 GiB	7.46 GiB
Q4_K_M	7.52 GiB	17.28 GiB
Licence	apache-2.0	apache-2.0

KV cache by context

the term that decides long-context viability

Context	Mellum2-12B-A2.5B-Thinking	Qwen3-Coder-30B-A3B-Instruct	Ratio
4,096	0.12 GiB	0.38 GiB	3.23×
8,192	0.17 GiB	0.75 GiB	4.39×
16,384	0.28 GiB	1.50 GiB	5.35×
32,768	0.50 GiB	3.00 GiB	6.01×
65,536	0.94 GiB	6.00 GiB	6.41×
131,072	1.81 GiB	12.00 GiB	6.62×