text-embedding-3-small
OpenAI's budget embedding workhorse: 5x cheaper than ada-002 with better multilingual quality and adjustable dimensions.
About
text-embedding-3-small is the efficiency play in OpenAI's third-generation embedding pair, launched in January 2024 alongside the large variant. It keeps the same 8,191-token input limit and 1,536 default dimensions as ada-002 but adds native dimension shortening through the dimensions parameter, so you can trim vectors to 512 or 256 for cheaper storage without retraining anything.
On OpenAI's own numbers it scores 62.3% average on MTEB English tasks and 44.0% on the MIRACL multilingual retrieval set, beating ada-002 on both while costing a fifth as much. It remains the default pick for teams that live on the OpenAI API and want retrieval-grade embeddings at prototype pricing.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- MTEB English average: 62.3% (provider-reported, OpenAI model card)
- MIRACL multilingual retrieval average: 44.0% vs 31.4% for ada-002 (provider-reported)
- BEIR retrieval nDCG@10: ~51-53% (independent mid-2025 field comparison)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: official OpenAI launch post and model card publish the 62.3% MTEB and 44.0% MIRACL figures; pricing confirmed on OpenAI's pricing page and in independent 2026 comparisons. Strengths: massive production adoption, stable API since Jan 2024, dimension control without retraining. Docked for: (1) weights are closed, so the headline figures cannot be independently reproduced; (2) no free tier.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 10.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 3.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · MTEB 62.3% and MIRACL 44.0% from OpenAI's own model card; pricing $0.02/1M confirmed on OpenAI pricing page
- · Released 2024-01-25; stable production API with Batch API discount ($0.01/1M)
- · No free tier; dimension shortening via the dimensions parameter verified in docs