Global Araç
Embeddings Cost Comparison
| Sağlayıcı | Boyut | MTEB | $/1M token | Aylık |
|---|---|---|---|---|
| BGE-M3 (self-hosted) | 1024 | 59 | $0.00 | $0 |
| OpenAI text-embedding-3-small | 1536 | 62.3 | $0.02 | $1.6 |
| Voyage 3 lite | 512 | 60.8 | $0.02 | $1.6 |
| Gemini text-embedding-005 | 768 | 61.5 | $0.03 | $2 |
| Mistral embed | 1024 | 60 | $0.10 | $8 |
| Cohere embed-v4 | 1536 | 63.9 | $0.12 | $9.6 |
| OpenAI text-embedding-3-large | 3072 | 64.6 | $0.13 | $10.4 |
| Voyage 3 large | 1024 | 65.1 | $0.18 | $14.4 |
MTEB is the standard public retrieval benchmark (avg of 56 tasks). 1-2 points of MTEB rarely matters for typical RAG; 5+ points does. Cheap models like text-embedding-3-small and BGE-M3 usually win the cost/quality tradeoff — reserve text-embedding-3-large or Voyage 3 large for niche domains where retrieval recall is mission-critical.
Eight embedding providers compared by cost per million tokens, vector dimensions, and MTEB benchmark scores (the standard public retrieval benchmark, average of 56 tasks). 1-2 points of MTEB rarely matters for typical RAG; 5+ does. Cheap models like text-embedding-3-small and BGE-M3 usually win the cost/quality tradeoff.
Nasıl Kullanılır
- Enter document count + tokens per doc.
- Enter monthly re-embeds.
- Read sorted cost table.