MIRACL average (multi-language retrieval)
We haven't written a description for this benchmark yet; see the card notes below for how each maker ran it.
- ModelScoreYour cost / mo
- 1text-embedding-3-large · OpenAI54.9%
- 2text-embedding-3-small · OpenAI44.0%
“Best value”: no other model here reports a higher score at a lower cost for your workload. Scores are self-reported by each maker with different settings, so treat close calls as ties.
How each maker reported it
- text-embedding-3-large launch post (eval table); embeddings guide
OpenAI's own numbers from the January 2024 launch post: MIRACL = multi-language retrieval average, MTEB = English tasks average. OpenAI also states a 3-large vector shortened to 256 dimensions still beats an unshortened ada-002 (1536) on MTEB.
- text-embedding-3-small launch post (eval table); embeddings guide
OpenAI's own numbers from the January 2024 launch post: MIRACL = multi-language retrieval average, MTEB = English tasks average. OpenAI also states a 3-large vector shortened to 256 dimensions still beats an unshortened ada-002 (1536) on MTEB.
Only each model's own maker's number is shown; scores a maker reports for rival models are left out.