Choosing an AI API for RAG (retrieval-augmented generation)

What RAG (retrieval-augmented generation) needs

• Low input price for large retrieved context
• Faithfulness to the supplied passages
• Citations back to sources

Models that fit

Somnus column uses the minimum 2× credit; larger top-ups go further.

Cost example

1,000 requests of about 1,500 input and 400 output tokens on Gemini 3.8 Flash cost about $2.63 at list price — roughly $1.31 with Somnus credit.

Tips

• Retrieve fewer, better passages rather than many weak ones.
• Number the passages and ask the model to cite them.
• Tell the model to answer 'not in the sources' when appropriate.

Common mistakes

• Sending the top 50 chunks when the top 5 would do.
• Not instructing the model to refuse when context is missing.

Example

from openai import OpenAI
import os

client = OpenAI(
    api_key=os.environ["SOMNUS_API_KEY"],
    base_url="https://gateway-production-c837.up.railway.app/v1",
)

resp = client.chat.completions.create(
    model="claude-sonnet-5",
    messages=[
      {"role": "system", "content": "You are an expert assistant. Be precise and concise."},
      {"role": "user", "content": "Using only the passages below, when was the company founded?"}
    ],
)
print(resp.choices[0].message.content)

Questions

Related