Model comparison
Nanonets-OCR-s vs Qwen3-VL-2B-Instruct
At Q4_K_M, Qwen3-VL-2B-Instruct is the smaller download — 1,107,409,952 bytes against 1,929,900,800. At long context the gap widens: Nanonets-OCR-s's KV cache at 32K is 3.1× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Nanonets-OCR-s | Qwen3-VL-2B-Instruct | |
|---|---|---|
| Parameters | 3.8B | 2.1B |
| Architecture | qwen2vl | qwen3vl |
| Layers | 36 | 28 |
| Native context | 128,000 | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 56 | 46 |
| Smallest quantization | 0.74 GiB | 0.50 GiB |
| Q4_K_M | 1.80 GiB | 1.03 GiB |
| Licence | — | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Nanonets-OCR-s | Qwen3-VL-2B-Instruct | Ratio |
|---|---|---|---|
| 4,096 | 0.14 GiB | 0.44 GiB | 3.11× |
| 8,192 | 0.28 GiB | 0.88 GiB | 3.11× |
| 16,384 | 0.56 GiB | 1.75 GiB | 3.11× |
| 32,768 | 1.13 GiB | 3.50 GiB | 3.11× |
| 65,536 | 2.25 GiB | 7.00 GiB | 3.11× |
| 131,072 | 4.50 GiB | 14.00 GiB | 3.11× |