MODELS
text-embedding-3-large
OpenAI's flagship embedding model for RAG and semantic search.
Specs
- Context window
- 8,191
- Modalities
- text
- Tool use
- —
- Vision
- —
- Streaming
- —
- License
- proprietary
- Released
- 2024-01-25
Pricing
- Input / 1M
- $0.13
text-embedding-3-large (January 2024) is OpenAI's top-tier embedding model — 3072 dimensions native, with Matryoshka-style truncation to 256/512/1024 dims for cost-quality trade-offs. $0.13 per million tokens, 8K input window. Strong on multilingual retrieval, including Mandarin/Cantonese/traditional Chinese. The default RAG building block for OpenAI-stack apps.
Editor's verdict
Solid default — but be aware of cheaper alternatives that are now competitive. BGE-M3 (open-source, multilingual, free to self-host) and Cohere Embed v3 both match or beat it on Chinese retrieval. For pure English, the OpenAI smaller variant text-embedding-3-small at $0.02/M is fine for most apps. Pick text-embedding-3-large when you want zero-config quality on a known stack and cost is not the constraint.
Reviews
No reviews yet. Be the first.
Last updated: 2026-04-29