The cheapest AI model router

What the same model costs, side by side

USD per 1,000,000 tokens, read live from the catalogue. "Buying direct" is the provider's own published rate for the identical model. Every model on the full catalogue carries the same 30% reduction.

Why cheaper is worth acting on here

A discount only matters if switching costs less than it saves. Here it costs two lines. The API implements the OpenAI /v1/chat/completions interface, so the official OpenAI SDKs keep working: you change the base URL and the key, and the model string picks which of the 458 models runs.

from openai import OpenAI

client = OpenAI(
    base_url="https://tokenroute.app/api/v1",   # was https://api.openai.com/v1
    api_key="sk-tr-...",                  # your TokenRoute key
)

# Everything below is unchanged.
response = client.chat.completions.create(
    model="openai/gpt-5",
    messages=[{"role": "user", "content": "Hello!"}],
)

What else the price does not hide

• No subscription, seat fee or minimum spend. You are billed for the tokens a request actually consumed, and nothing else.
• 21 models cost nothing at all. They run on a zero balance, with no card on file.
• Per-key spend limits. Cap what any single API key can spend and revoke it instantly, so a runaway loop cannot drain a balance.
• Credits never expire on a cycle. A balance you top up stays yours until requests consume it, with no monthly reset.

Questions people ask before switching

Is it the same model?

Yes. "openai/gpt-5" is GPT-5, not a distilled or quantised substitute. The model string names the provider and the model, and that is what runs.

Is the discount permanent?

The 70% rate is applied to every model in the catalogue and is the published price, not an introductory one. Prices track each provider’s own rate, so when they cut theirs, yours falls with it.

Do I need a credit card to start?

No. Creating an account and generating keys is free, and the 21 free models run on a zero balance. You only need a payment method to call a paid model.

What happens to my existing code?

Nothing, as long as it speaks the OpenAI interface. Request and response shapes, streaming over Server-Sent Events, and tool calling are unchanged, which is why LangChain, LlamaIndex and the Vercel AI SDK work as they are.