Qwen 3 Embedding 4B vs Qwen 3 Embedding 8B
Qwen 3 Embedding 4B69.4
Qwen 3 Embedding 8B70.6
| Benchmark | Qwen 3 Embedding 4B | Qwen 3 Embedding 8B |
|---|---|---|
| MTEB multilingual v2 · Mean over tasks · multilingual (1037 languages) | 69.4 | 70.6 |
Specs
| Spec | Qwen 3 Embedding 4B | Qwen 3 Embedding 8B |
|---|---|---|
| Price | !$0.02 | !$0.01 |
| Context | 33K | 33K |
| Weights | Open | Open |
| Released | Jun 2025 | Jun 2025 |
| Published speed | — | 3.3 × vLLM H100 (Baseten) |
| Dimensions | 2,560 | 4,096 |
| Max input | 33K tokens | 33K tokens |
| Memory (4-bit) | 2.4 GB | 4.5 GB |
| Licence | Apache-2.0 | Apache-2.0 |
| Commercial use | Allowed | Allowed |
What breaks when you switch
Qwen 3 Embedding 4B → Qwen 3 Embedding 8B
Re-embed your whole corpus›
Vectors from different models can't be mixed; dimensions change from 2,560 to 4,096.
Memory (4-bit): 2.4 GB → 4.5 GB
Qwen 3 Embedding 8B → Qwen 3 Embedding 4B
Re-embed your whole corpus›
Vectors from different models can't be mixed; dimensions change from 4,096 to 2,560.