Llama Embed Nemotron 8B vs Qwen 3 Embedding 4B
Llama Embed Nemotron 8B69.5
Qwen 3 Embedding 4B69.4
| Benchmark | Llama Embed Nemotron 8B | Qwen 3 Embedding 4B |
|---|---|---|
| MTEB multilingual v2 · Mean over tasks · multilingual (1037 languages) | 69.5 | 69.4 |
Specs
| Spec | Llama Embed Nemotron 8B | Qwen 3 Embedding 4B |
|---|---|---|
| Price | — | !$0.02 |
| Context | 33K | 33K |
| Weights | Open | Open |
| Released | Oct 2025 | Jun 2025 |
| Dimensions | 4,096 | 2,560 |
| Max input | 131K tokens | 33K tokens |
| Memory (4-bit) | 4.5 GB | 2.4 GB |
| Licence | https://huggingface.co/nvidia/llama-embed-nemotron-8b/blob/main/LICENSE | Apache-2.0 |
| Commercial use | Restricted | Allowed |
What breaks when you switch
Llama Embed Nemotron 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.
Max input: 131K tokens → 33K tokens
Qwen 3 Embedding 4B → Llama Embed Nemotron 8B
Re-embed your whole corpus›
Vectors from different models can't be mixed; dimensions change from 2,560 to 4,096.