Model comparison
JoyAI-Echo vs MiniMax-H3
At Q4_K_M, MiniMax-H3 is the smaller download — 14,576,977,960 bytes against 15,694,613,568.
From the file· summed bytes, KV per layer
Side by side
| JoyAI-Echo | MiniMax-H3 | |
|---|---|---|
| Parameters | 12.2B | 33.1B |
| Architecture | joyai-gemma | qwen3vl |
| Layers | — | — |
| Native context | — | — |
| Mixture of experts | no | no |
| Quantizations published | 5 | 22 |
| Smallest quantization | 8.54 GiB | 10.60 GiB |
| Q4_K_M | 14.62 GiB | 13.58 GiB |
| Licence | — | other |
KV cache by context
the term that decides long-context viability
| Context | JoyAI-Echo | MiniMax-H3 | Ratio |
|---|---|---|---|
| 4,096 | — | — | — |
| 8,192 | — | — | — |
| 16,384 | — | — | — |
| 32,768 | — | — | — |
| 65,536 | — | — | — |
| 131,072 | — | — | — |