Model comparison
Qwen3-Embedding-8B vs Qwen3-Embedding-4B
At Q4_K_M, Qwen3-Embedding-4B is the smaller download — 2,496,704,224 bytes against 4,676,804,928.
From the file· summed bytes, KV per layer
Side by side
| Qwen3-Embedding-8B | Qwen3-Embedding-4B | |
|---|---|---|
| Parameters | 7.6B | 4.0B |
| Architecture | qwen3 | qwen3 |
| Layers | 36 | 36 |
| Native context | 40,960 | 40,960 |
| Mixture of experts | no | no |
| Quantizations published | 17 | 18 |
| Smallest quantization | 2.87 GiB | 1.55 GiB |
| Q4_K_M | 4.36 GiB | 2.33 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3-Embedding-8B | Qwen3-Embedding-4B | Ratio |
|---|---|---|---|
| 4,096 | 0.56 GiB | 0.56 GiB | 1.00× |
| 8,192 | 1.13 GiB | 1.13 GiB | 1.00× |
| 16,384 | 2.25 GiB | 2.25 GiB | 1.00× |
| 32,768 | 4.50 GiB | 4.50 GiB | 1.00× |
| 65,536 | 9.00 GiB | 9.00 GiB | 1.00× |
| 131,072 | 18.00 GiB | 18.00 GiB | 1.00× |