← Back to all frameworks Graph RAG

text-embedding-3 / Voyage / Cohere

Best-in-class embedding models for retrieval quality

What it is

OpenAI text-embedding-3, Voyage AI, Cohere embed-v3 — the current top-of-class for English and multilingual semantic encoding. The single biggest lever on RAG quality.

How Vaaani uses it

  • Encoding documents into a vector DB
  • Multilingual semantic search across mixed languages
  • Code embeddings (Voyage code) for codebase search
  • Domain-tuned variants for legal / medical / financial corpora

Why it makes the cut

Most RAG quality issues are embedding quality issues. Switching from a free model to text-embedding-3-large often beats spending a week on prompt engineering.

Sample code

from openai import OpenAI
client = OpenAI()

emb = client.embeddings.create(
    model="text-embedding-3-large",
    input=["Vaaani builds AI workers"])
vec = emb.data[0].embedding   # 3072-dim

Related in the Vaaani stack

Have a project that needs text-embedding-3?

30-min discovery call. You describe the busywork; I map it to an AI worker and a budget.