text-embedding-3-large

Embeddings turn text into lists of numbers that capture its meaning, so you can search by meaning rather than keywords, find similar documents, or let an assistant draw on your own files. text-embedding-3-large is OpenAI’s most accurate: 3,072 numbers per text by default, scoring 64.6% on the MTEB benchmark.

Its real advantage is outside English. On MIRACL, a multilingual search test, it scores 54.9% against 44.0% for the small version. You can also shorten its vectors to save storage — OpenAI says a 256-number version still beats its older ada-002 model.

Every rate is published and stays the same from one request to the next. See every model.

text-embedding-3-large is billed in rupees, on the same monthly invoice as every other model your team uses — no foreign card, no separate OpenAI account, and one key for all of it.

Prompts sent to text-embedding-3-large are processed outside Sri Lanka. If a workload has to stay in the country, Qwen3.8-27B (Colombo) runs on our own servers in Colombo.

If your documents mix English with Sinhala or Tamil, this is the better of OpenAI’s two embedding models for multilingual search. OpenAI publishes no Sinhala or Tamil results, though, so check retrieval quality on your own documents.

Use the OpenAI SDK you already have — point it at Roar AI and name the model. Read the docs.

from openai import OpenAI

client = OpenAI(base_url="https://api.roar-ai.com/v1", api_key="roar_live_…")

vectors = client.embeddings.create(
    model="text-embedding-3-large",
    input=["Quarterly sales report, western province"],
)