Best Embedding Model: Compare Price, Dimensions & MTEB Quality

Choosing a text-embedding model comes down to four numbers: how much it costs per token, how good its vectors are at retrieval, how many tokens it accepts per call, and how large each output vector is. This tool puts all four side by side for 15 of the most-used models and then projects the cost to embed your own corpus.

The cost projection uses one formula:

1. Total tokens = number of documents × average tokens per document.
2. Embedding cost = (total tokens ÷ 1,000,000) × price per 1M tokens. Each model's price comes from its provider's public pricing page, cited below.
3. Open-weight models(bge, gte, e5, nomic) show $0 API cost with a “self-host” label. We deliberately do not invent a GPU/compute price — that depends on your hardware — and link to the self-hosting cost calculator instead.
4. Monthly re-embedding multiplies the one-time cost by 12 to show annualised spend; one-time shows the single figure.

Quality is the MTEB (Massive Text Embedding Benchmark) average, displayed verbatim from the leaderboard snapshot — never computed by this page. The MTEB(eng, v1) — snapshot 2026-06-20is used so every score is comparable; the Retrieval sub-score is the most relevant one for search and RAG. Where a provider only publishes private or multilingual benchmarks, the MTEB cell shows “n/a” and that model is excluded from the quality ranking rather than guessed at.

The three recommendation cards are deterministic. Cheapest is the lowest projected cost. Best quality is the highest MTEB average. Best value maximises MTEB ÷ max(cost, $0.01) — quality per dollar — with the one-cent floor letting free self-hosted models rank by their raw MTEB instead of dividing by zero. Because all rates and scores are constants, identical inputs always produce identical output. The per-document and bulk cost formulas are cross-checked against each other in the data module to guard against arithmetic drift.

• MTEB Leaderboard — quality scores (Average + Retrieval)
• OpenAI — new embedding models & MTEB scores
• OpenAI — API pricing (per-1M-token embedding rates)
• Cohere — Embed v3 announcement & benchmarks
• Voyage AI — embeddings docs & pricing
• Google — Gemini API embeddings
• Mistral — embeddings capability & pricing
• Hugging Face — open model cards (bge, gte, e5, nomic)

Prices and MTEB scores on this page are a static snapshot last verified on 2026-06-20. Embedding pricing and the leaderboard change often — this page is reviewed quarterly. Spotted a stale number? Email me and I'll update it.