Use the Vercel AI SDK with UberLLM (cheaper models, same code)

Steps

1. Make a key on your keys page and add a little credit (accounts start at $0).
2. Install the provider:

npm i ai @ai-sdk/openai-compatible

3. Put the key in .env.local (Next.js) or your server's environment:

UBERLLM_API_KEY=ull_your_key

4. Create the provider once, for example in lib/uberllm.ts:

import { createOpenAICompatible } from "@ai-sdk/openai-compatible";

export const uberllm = createOpenAICompatible({
  name: "uberllm",
  baseURL: "https://api.uberllm.dev/v1",
  apiKey: process.env.UBERLLM_API_KEY, // sent as Authorization: Bearer <key>
  includeUsage: true, // token counts on streamed answers too
});

5. Use it where you used a model before:

import { generateText } from "ai";
import { uberllm } from "@/lib/uberllm";

const { text } = await generateText({
  model: uberllm("deepseek/deepseek-v4.1-flash"),
  prompt: "Write a one-line welcome message.",
});

Moving an app that already uses @ai-sdk/openai

Swap the model argument and keep everything else:

// before
import { openai } from "@ai-sdk/openai";
const model = openai("gpt-4o");

// after
import { uberllm } from "@/lib/uberllm";
const model = uberllm("deepseek/deepseek-v4.1-flash");

Model ids are the ids on uberllm.dev/models and always have a maker in front (openai/…, anthropic/…, deepseek/…). That is how you can run GPT, Claude, Gemini and open models from one key and compare them by changing one string.

A Next.js chat route for useChat

useChat from @ai-sdk/react posts to /api/chat by default. This route answers it:

// app/api/chat/route.ts
import { streamText, convertToModelMessages, type UIMessage } from "ai";
import { uberllm } from "@/lib/uberllm";

export async function POST(req: Request) {
  const { messages }: { messages: UIMessage[] } = await req.json();
  const result = streamText({
    model: uberllm("deepseek/deepseek-v4.1-flash"),
    messages: await convertToModelMessages(messages),
  });
  return result.toUIMessageStreamResponse();
}

Keep the key on the server. Never put it in a NEXT_PUBLIC_ variable: anyone could read it from the page and spend your balance.

Cap what one call can cost

Chat requests accept provider.max_price (dollars per million tokens). Pass it through providerOptions under the provider name you chose:

await generateText({
  model: uberllm("deepseek/deepseek-v4.1-flash"),
  prompt: "Hi",
  providerOptions: { uberllm: { provider: { max_price: { prompt: 0.2, completion: 0.4 } } } },
});

If nobody sells the model under that price right now the call is refused with max_price_exceeded and nothing is charged.

Troubleshooting

• 401 invalid_api_key: the key did not reach the server. Restart next dev after editing .env.local, and check process.env.UBERLLM_API_KEY is set where the code runs (server, not the browser).
• 402 insufficient_credits: add credit on the credits page.
• 404 model_not_found: the id is wrong or missing its maker. Copy it from the model's page.
• 400 wrong_endpoint: you passed an image or video model to streamText. Image models go through uberllm.imageModel(…) with generateImage; video is a job, see Make a video or image.
• Usage is empty on streams: set includeUsage: true on the provider, as above.

Keep reading

• The full reference with tool calls and images: Vercel AI SDK doc
• OpenAI-compatible endpoint explained
• Live LLM prices