Frontier models, up to 80% under list.

Claude, GPT, Gemini, DeepSeek, Kimi and more through one OpenAI- and Anthropic-compatible API. Pay as you go from $5.

Claude Code with its API base URL set to https://api.routing.run answers a request, and the dashboard shows what it cost: $0.0367 on Claude Opus 5, $0.122 at list prices, 70% less.

Start in one step

Point your client at routing.run. The rest of your code stays as it is.

import osfrom openai import OpenAIclient = OpenAI(    base_url="https://api.routing.run/v1",    api_key=os.environ["ROUTING_API_KEY"],)client.chat.completions.create(    model="claude-opus-5",    messages=[{"role": "user", "content": "Hi"}],)

Change two arguments and keep the rest. Chat completions, responses and the model list all work, for every model.

Get a key

What you get

29 models, one key. 12 labs on one balance, at the prices in the table below.

See the models

Pay per token. Each request comes off your credits.

Top up from$5Credits you buy don't expire. No plan, no seats.

A key for each app. See when each one was last used, and revoke it the moment it leaks.

Choose a model family

Each opens the price table below, filtered to that family.

Models and prices

Every model next to its list price. What you see is what each token costs.

29 models
Models with their prices per million tokens, ours and at list
Claude Fable 5.2$5.60list $8.00$28.00list $40.001M30%
Claude Opus 5.5$2.80list $4.00$14.00list $20.001M30%
Claude Sonnet 5.5$1.80list $2.00$9.00list $10.001M10%
Claude Haiku 5.5$0.07list $0.10$0.35list $0.50100K30%
Claude Opus 5$1.50list $5.00$7.50list $25.001M70%
Claude Sonnet 5$0.80list $2.00$4.00list $10.001M60%
GPT-6 Astra$8.00list $10.00$40.00list $50.001.05M20%
GPT-6.1 Sol$1.80list $2.00$9.00list $10.001.05M10%
GPT-6.1 Terra$0.90list $1.00$4.50list $5.001.05M10%
GPT-6 Luna$0.095list $0.10$0.475list $0.501.05M5%
GPT-OSS 120B$0.014list $0.07$0.06list $0.30128K80%
Gemini 4 Argon$0.50list $2.00$2.50list $10.001.05M75%

Prices per million tokens. Ours rounded up, list rounded down; the exact figures are at api.routing.run/v1/pricing. Select an id to copy it.

Works with

Anything that speaks the OpenAI or Anthropic API. Change one setting and add your key.

What you'd save

Pay as you go from $5, no plan and no monthly fee. The figures use the per-token prices you pay.

Input to output
120M

You save each month

$201.60

$2,419 a year, 30% under list

At list
$672.00
On routing.run
$470.40

108M input and 12M output tokens at Claude Opus 5.5's per-token prices. Cached input costs less again.

How billing works

Credits in, tokens out. No plan, no seats, no monthly fee.

  1. 1. Add credits

    Sign in with GitHub, Google, Discord, Apple or an email link, then top up from $5. Credits you buy don't expire.

    Open the dashboard
  2. 2. Create a key

    Make one key for each place your code runs and name it after that place. The dashboard shows when each key was last used, and you can revoke one the moment it leaks. This call lists every model:

    curl https://api.routing.run/v1/models \  -H "Authorization: Bearer $ROUTING_API_KEY"
  3. 3. Pay per token

    Each request is billed per token at the price in the table above and taken from your credits. Usage lists every request with its model, tokens and cost. We don't keep prompts or answers.

Questions

Anything else: support@routing.run

What do I pay?

You pay less than list for the same model id, at the price in the table. That price is what you pay per token.

Do I get the model I ask for?

Yes. We send each request to the model id you ask for and never substitute another model. The table shows the context each one accepts.

What works with it?

Anything that speaks the OpenAI or Anthropic API: the official SDKs, LangChain and similar libraries, and coding agents such as Claude Code, Codex, OpenCode and Cline. Use https://api.routing.run/v1 as the OpenAI base URL, or https://api.routing.run for Anthropic clients.

Do you store my prompts?

We don't store the content of your prompts or responses. They pass through memory to the model and are discarded once the response is sent. We keep request metadata (model, token counts, cost, timing) for billing.

What happens if a model is unavailable?

The API answers 503 with a short message. Retry after a moment, or send the request to another model id.

Do credits expire?

Credits you buy don't expire. The smallest top-up is $5.

How do I sign in?

With GitHub, Google, Discord or Apple, or with a sign-in link sent to your email. There is no password to set.