Embedding Model Comparison Table

Side-by-side comparison of every major embedding model: dimensions, max input tokens, price per million tokens, MTEB benchmark score, and whether it's open-weight. Filter, sort, and click any model for the full spec card.

How to use the Embedding Model Comparison Table

Sort by MTEB score (overall retrieval quality), price (cheapest first), embedding dimensions (smaller = cheaper to store), or max input. Filter to open-weight models if you want to self-host. MTEB scores are the official aggregated leaderboard average — higher is better.

How to pick an embedding model

Three factors: quality (MTEB), price (for both initial corpus embedding and per-query), dimensions (storage cost in your vector DB scales linearly).

For most production RAG: start with text-embedding-3-small (cheap, 1536-dim, MTEB ~62) or voyage-3-lite. Move to text-embedding-3-large or voyage-3 if retrieval quality is your bottleneck. For multilingual: multilingual-e5-large open-weight, or cohere-embed-multilingual-v3.

The 256-dim variants (where supported via dimensions param) cut storage 6x at modest quality cost — worth measuring on your data.

Frequently asked questions

What does the MTEB score tell me?
MTEB is a broad benchmark averaging many retrieval and classification tasks. A higher score suggests stronger general-purpose quality, but the best model still depends on your specific data.

Do more dimensions mean a better model?
Not necessarily. Larger vectors can capture more but cost more to store and compare. Several smaller models match or beat larger ones on MTEB.

How current are the numbers?
The table reflects published specifications and prices at the last update. Vendors change pricing, so confirm against the provider before committing.
Embed this tool on your site

Free to embed, no attribution required (but appreciated). Paste this where you want the tool to appear: