Model comparison
Qwen3-Embedding-8B vs Nemotron-3-Embed-8B-BF16
At Q4_K_M, Qwen3-Embedding-8B is the smaller download — 4,676,804,928 bytes against 4,896,389,984.
From the file· summed bytes, KV per layer
Side by side
| Qwen3-Embedding-8B | Nemotron-3-Embed-8B-BF16 | |
|---|---|---|
| Parameters | 7.6B | 8.0B |
| Architecture | qwen3 | mistral3 |
| Layers | 36 | 34 |
| Native context | 40,960 | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 17 | 36 |
| Smallest quantization | 2.87 GiB | 1.81 GiB |
| Q4_K_M | 4.36 GiB | 4.56 GiB |
| Licence | apache-2.0 | openmdw-1.1 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3-Embedding-8B | Nemotron-3-Embed-8B-BF16 | Ratio |
|---|---|---|---|
| 4,096 | 0.56 GiB | 0.53 GiB | 1.06× |
| 8,192 | 1.13 GiB | 1.06 GiB | 1.06× |
| 16,384 | 2.25 GiB | 2.13 GiB | 1.06× |
| 32,768 | 4.50 GiB | 4.25 GiB | 1.06× |
| 65,536 | 9.00 GiB | 8.50 GiB | 1.06× |
| 131,072 | 18.00 GiB | 17.00 GiB | 1.06× |