AI Embedding API Benchmark: 8 Providers

Embedding quality, latency, cost across 8 APIs. With retrieval accuracy tests.

What we measured

8 embedding APIs (OpenAI, Cohere, Voyage, Jina, Google, Mistral, DeepSeek, ada-002). 1000 query-passage pairs, MTEB subset.

Quality results

NDCG@10 by provider and model size. Cost per million tokens. Latency p50/p99.

Retrieval test

End-to-end RAG accuracy. Synthetic Q&A dataset, 200 questions, 5000 context chunks.

Code

Reproducible harness in bench/. Includes baseline BM25 for comparison.

Read more

AI API cost analysis: codecost.apiguider.com. All benchmarks: apibenchmarks.apiguider.com.

---

Part of the apiguider.com network:

• Apex (editorial home): https://apex.apiguider.com
• All 22 stations: https://apex.apiguider.com/22-stations/
• Editorial standards: https://apex.apiguider.com/editorial-standards/

Author: apiguider.com editorial team (see apex.apiguider.com/about/)

Topics: ai-api, embeddings, rag, retrieval