Model comparison
medgemma-4b-it vs Qwen3-VL-2B-Instruct
At Q4_K_M, Qwen3-VL-2B-Instruct is the smaller download — 1,107,409,952 bytes against 2,489,893,664. At long context the gap widens: medgemma-4b-it's KV cache at 32K is 4.4× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| medgemma-4b-it | Qwen3-VL-2B-Instruct | |
|---|---|---|
| Parameters | 4.3B | 2.1B |
| Architecture | gemma3 | qwen3vl |
| Layers | 34 | 28 |
| Native context | 131,072 | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 46 | 46 |
| Smallest quantization | 1.10 GiB | 0.50 GiB |
| Q4_K_M | 2.32 GiB | 1.03 GiB |
| Licence | other | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | medgemma-4b-it | Qwen3-VL-2B-Instruct | Ratio |
|---|---|---|---|
| 4,096 | 0.25 GiB | 0.44 GiB | 1.76× |
| 8,192 | 0.33 GiB | 0.88 GiB | 2.68× |
| 16,384 | 0.48 GiB | 1.75 GiB | 3.63× |
| 32,768 | 0.79 GiB | 3.50 GiB | 4.40× |
| 65,536 | 1.42 GiB | 7.00 GiB | 4.93× |
| 131,072 | 2.67 GiB | 14.00 GiB | 5.24× |