Embedding Model API Pricing 2026: OpenAI vs Cohere vs Voyage vs Open Source

Embedding model cost is the silent budget killer in production AI systems. A pipeline processing 500K documents can cost hundreds per month. ## Commercial Embedding APIs ### OpenAI text-embedding-3-small - Price: $0.02 per 1M tokens - Dimensions: 512 - Cost for 1M docs: $10 ### OpenAI text-embedding-3-large - Price: $0.13 per 1M tokens - Dimensions: 3,072 - Cost for 1M docs: $65 ### Cohere Embed v3 - Paid: $0.10 per 1K embeddings - Dimensions: 1,024 - Best for: Multilingual (100+ languages) ### Voyage AI - Price: ~$0.03 per 1M tokens - Dimensions: 1,024 - Best for: Code-heavy documents ## Open Source (Self-Host) | Model | Size | Dimensions | GPU | |-------|------|-----------|-----| | bge-small-en-v1.5 | 33MB | 384 | CPU OK | | bge-base-en-v1.5 | 110MB | 768 | CPU OK | | bge-large-en-v1.5 | 335MB | 1,024 | GPU nice | | e5-mistral-7b | 14GB | 4,096 | GPU required | ## Cost Comparison (1M Documents) | Provider | Cost | Speed | Quality | |----------|------|-------|---------| | OpenAI 3-small | $10 | Fast | High | | OpenAI 3-large | $65 | Fast | Highest | | Cohere | ~$100 | Fast | High | | Voyage | ~$3 | Fast | High | | bge-small (self-host) | $0 + GPU | Medium | Medium | | bge-large (self-host) | $0 + GPU | Medium | High | ## When to Use What - Under 10K docs: OpenAI 3-small - 10K-100K docs: Voyage (best price-to-quality ratio) - 100K-1M docs: Self-host bge-base on single GPU - Over 1M docs: Self-host bge-large ## The Hidden Cost: Latency OpenAI API: 150-300ms per batch. Self-hosted bge-small on T4: ~5ms per batch. For 1M docs: 41 hours vs 1.4 hours. ## Bottom Line Best setup: bge-base self-hosted for production (>100K docs), OpenAI 3-small for prototyping. All pricing from official sources as of July 2026.