Giga Embeddings Instruct 3B 0826 vs Giga Embeddings Instruct 480M 0826
Giga Embeddings Instruct 3B 082663.9
Giga Embeddings Instruct 480M 0826—
| Benchmark | Giga Embeddings Instruct 3B 0826 | Giga Embeddings Instruct 480M 0826 |
|---|---|---|
| MTEB multilingual v2 · Mean over tasks · multilingual (1037 languages) | 63.9 | — |
Specs
| Spec | Giga Embeddings Instruct 3B 0826 | Giga Embeddings Instruct 480M 0826 |
|---|---|---|
| Context | 8K | 8K |
| Weights | Open | Open |
| Released | Aug 2026 | Aug 2026 |
| Dimensions | 2,048 | 1,024 |
| Max input | 8K tokens | 8K tokens |
| Languages | en, ru | en, ru |
| Memory (4-bit) | 1.9 GB | 0.3 GB |
| Licence | MIT | MIT |
| Commercial use | Allowed | Allowed |
What breaks when you switch
Giga Embeddings Instruct 3B 0826 → Giga Embeddings Instruct 480M 0826
Re-embed your whole corpus›
Vectors from different models can't be mixed; dimensions change from 2,048 to 1,024.
Giga Embeddings Instruct 480M 0826 → Giga Embeddings Instruct 3B 0826
Re-embed your whole corpus›
Vectors from different models can't be mixed; dimensions change from 1,024 to 2,048.