Use the OpenAI Python SDK and LangChain with a cheaper API (UberLLM)

Steps (OpenAI Python SDK)

1. Make a key on your keys page and add credit.
2. Save it as an environment variable:

export UBERLLM_API_KEY=ull_your_key

3. Point the client at UberLLM:

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.uberllm.dev/v1",
    api_key=os.environ["UBERLLM_API_KEY"],
)

resp = client.chat.completions.create(
    model="deepseek/deepseek-v4.1-flash",
    messages=[{"role": "user", "content": "Give me three names for a coffee shop."}],
)
print(resp.choices[0].message.content)

That is the whole change. client.chat.completions.create with stream=True, tools, response_format, temperature and max_tokens works as before. The async client (AsyncOpenAI) takes the same two arguments.

Without touching the code

The OpenAI SDK also reads two environment variables, so a script that calls OpenAI() with no arguments can be moved without editing it:

export OPENAI_BASE_URL=https://api.uberllm.dev/v1
export OPENAI_API_KEY=ull_your_key
python your_script.py

You still need to change the model= string to an UberLLM id such as deepseek/deepseek-v4.1-flash.

Steps (LangChain)

pip install langchain-openai

import os
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="deepseek/deepseek-v4.1-flash",
    base_url="https://api.uberllm.dev/v1",
    api_key=os.environ["UBERLLM_API_KEY"],
)
print(llm.invoke("Say hello in one sentence.").content)

Chains, agents, bind_tools and .stream() work unchanged. Pass base_url in code rather than relying on environment variables: the argument always wins, so a stray OPENAI_API_BASE on the machine cannot send calls elsewhere.

LangChain.js is the same idea:

import { ChatOpenAI } from "@langchain/openai";

const llm = new ChatOpenAI({
  model: "deepseek/deepseek-v4.1-flash",
  apiKey: process.env.UBERLLM_API_KEY,
  configuration: { baseURL: "https://api.uberllm.dev/v1" },
});

Pick a cheaper model for each job

The point of one key for every model is that you can stop paying frontier prices for easy work. A common split:

• Hard reasoning or customer-facing answers: a top model such as deepseek/deepseek-v4.1-flash.
• Bulk summaries, tagging, extraction: google/gemma-3-27b-it or another model from the batch table below, often 50 to 100 times cheaper per token.

Change only the model= string to move a step from one to the other.

What is not there

UberLLM serves chat completions (/v1/chat/completions), the Responses API (client.responses.create, streaming and tool calls included) and the Anthropic messages route. It keeps no state between calls, so send the whole conversation in input instead of previous_response_id (Responses API reference). It does not serve embeddings or the Assistants API: keep those calls on their current provider.

Troubleshooting

• openai.AuthenticationError (401): the key is wrong or empty. Print len(os.environ["UBERLLM_API_KEY"]); a key is ull_ plus 43 characters.
• openai.NotFoundError (404) on /responses or /embeddings: those routes do not exist here, see above.
• model_not_found: OpenAI-style ids like gpt-4o need the maker in front: openai/gpt-4o.
• openai.RateLimitError (429): the key's per-minute limit (300 by default). Raise it on the keys page, or add retries with backoff.
• 402 insufficient_credits: top up on the credits page.

Keep reading

• OpenAI SDK reference and LangChain doc
• The cheapest GPT API
• One API key for all LLMs