Both providers, one key

How routing works

One key reaches both providers. We route on the request path, not on anything you have to configure:

• /v1/messages → the Anthropic Messages API.
• /v1/chat/completions and /v1/responses → OpenAI.
• /v1/embeddings → OpenAI.

This means you keep using each provider’s official SDK in its native shape. We deliberately do not offer a unified abstraction layer over both. A lowest-common-denominator wrapper would cost you the features you are paying for, and there are good open-source options if you want one.

One key, two SDKs

import os
from anthropic import Anthropic
from openai import OpenAI

key = os.environ["LOWCOSTLLM_API_KEY"]

claude = Anthropic(base_url="https://api.lowcostllm.com", api_key=key)
gpt = OpenAI(base_url="https://api.lowcostllm.com/v1", api_key=key)

# Route by cost: cheap model first, escalate only when it matters.
draft = gpt.chat.completions.create(
    model="gpt-5.6-luna",
    messages=[{"role": "user", "content": prompt}],
)

review = claude.messages.create(
    model="claude-sonnet-5",
    max_tokens=2048,
    messages=[{"role": "user", "content": f"Critique this draft:\n{draft}"}],
)

Billing and limits

Spend across both providers draws from one balance and one set of tier limits. The usage endpoint groups by model, so a mixed workload still gives you a clean per-provider breakdown at the end of the month.