Model comparison
Omega-Evolution-26B-A4B-v3.0 vs Qwen3-30B-A3B-Thinking-2507
At Q4_K_M, Qwen3-30B-A3B-Thinking-2507 is the smaller download — 18,556,685,824 bytes against 51,258,890,720.
From the file· summed bytes, KV per layer
Side by side
| Omega-Evolution-26B-A4B-v3.0 | Qwen3-30B-A3B-Thinking-2507 | |
|---|---|---|
| Parameters | 26.5B | 30.5B |
| Architecture | gemma4 | qwen3moe |
| Layers | — | 48 |
| Native context | — | 262,144 |
| Mixture of experts | no | yes, 128 experts |
| Quantizations published | 5 | 51 |
| Smallest quantization | 13.10 GiB | 7.05 GiB |
| Q4_K_M | 47.74 GiB | 17.28 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Omega-Evolution-26B-A4B-v3.0 | Qwen3-30B-A3B-Thinking-2507 | Ratio |
|---|---|---|---|
| 4,096 | — | 0.38 GiB | — |
| 8,192 | — | 0.75 GiB | — |
| 16,384 | — | 1.50 GiB | — |
| 32,768 | — | 3.00 GiB | — |
| 65,536 | — | 6.00 GiB | — |
| 131,072 | — | 12.00 GiB | — |