Switch from OpenRouter to CheapTokens: base_url, key, model id

Your app already calls OpenRouter through the OpenAI SDK, and you want it to call CheapTokens instead. You change three values — base_url, the API key and the model id — and delete two optional headers. The rest of the code stays as it is.

Below are the exact lines for Python, JavaScript and curl, a quick way to confirm the switch worked, and a table of which model fits which job. If you have never sent a request to CheapTokens, the first API request guide covers the account and key; the docs have the full reference.

What actually changes#

OpenRouter's quickstart (checked 27 October 2026) configures the OpenAI client with the base URL https://openrouter.ai/api/v1 and names models as provider/model-name. CheapTokens speaks the same OpenAI Chat Completions format at https://api.cheaptokens.dev/v1. So the client class, chat.completions.create and the way you read the answer don't move.

That's one swap per line:

• base URL — https://api.cheaptokens.dev/v1 instead of the OpenRouter address;
• API key — a key from your CheapTokens account, read from an environment variable;
• model id — the full CheapTokens id, for example deepseek-ai/DeepSeek-V4-Flash-0731. Take ids from the CheapTokens docs: different providers name the same model differently, and short ids from price labels are not accepted.

And one deletion. OpenRouter's HTTP-Referer and X-OpenRouter-Title headers are optional and only feed OpenRouter's leaderboards. Remove them from default_headers / defaultHeaders; you don't need them here.

The code after the switch#

Same request, three runtimes. Pick the one your app uses.

import os
from openai import OpenAI

client = OpenAI(
    base_url='https://api.cheaptokens.dev/v1',
    api_key=os.environ['CHEAPTOKENS_API_KEY'],
)

resp = client.chat.completions.create(
    model='deepseek-ai/DeepSeek-V4-Flash-0731',
    messages=[{'role': 'user', 'content': 'Reply with one word: ready'}],
    max_tokens=20,
)
print(resp.choices[0].message.content)

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.cheaptokens.dev/v1',
  apiKey: process.env.CHEAPTOKENS_API_KEY,
});

const resp = await client.chat.completions.create({
  model: 'deepseek-ai/DeepSeek-V4-Flash-0731',
  messages: [{ role: 'user', content: 'Reply with one word: ready' }],
  max_tokens: 20,
});
console.log(resp.choices[0].message.content);

curl https://api.cheaptokens.dev/v1/chat/completions \
  -H "Authorization: Bearer $CHEAPTOKENS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek-ai/DeepSeek-V4-Flash-0731", "messages": [{"role": "user", "content": "Reply with one word: ready"}], "max_tokens": 20}'

One more thing before you run it. If your request carries OpenRouter-only fields — routing or fallback lists, provider preferences, ~alias model ids — those belong to OpenRouter. Keep the request to standard Chat Completions fields: model, messages, temperature, max_tokens, stream.

Check that it worked#

A reply isn't proof enough. Print three things next to it: which model answered, why it stopped, and what it billed in tokens.

print(resp.model)
print(resp.choices[0].finish_reason)
print(resp.usage.prompt_tokens, resp.usage.completion_tokens)

finish_reason of stop means the model finished on its own. length means max_tokens cut the answer off. usage shows input and output tokens separately, and those are the two numbers your bill is made of — the same two you were tracking on OpenRouter.

Which model replaces what#

Pick the model by the job, not by the name you used on OpenRouter. Facts below are from the CheapTokens price list checked 27 October 2026; one balance covers all three models and the price doesn't change with the time of day.

Not sure which row is yours? Start from what the app does:

What the same traffic costs#

Say your bot sends 1,000 requests, each with about 500 input tokens and 300 output tokens. That's an assumption — swap in your own numbers from usage. The table below is live: our prices and the OpenRouter column come from the current price list, input and output counted together.

On DeepSeek-V4-Flash those 1,000 requests come to $0.0375. Full prices are on the pricing page; the idea of what you're billed for is in input vs output tokens.

If it doesn't work#

After a switch, three errors cover most of the trouble. Match the message to the cause:

The 401 and 404 troubleshooting article goes through each one with the exact messages, and what base_url is explains why one wrong slash breaks the path.

Before you ship#