About TokenLab

One key for chat, image, video and audio models

TokenLab is an API and workspace for calling AI models from many makers through one account. It lists 432 models from 63 makers and bills each request from a prepaid balance.

Tell us what you need

Contact TokenLab

Need help?

Ask about an account or a failed request, or talk with us about pricing and partnerships.

Product support

Support requests

[email protected]

Business

Partnerships and sales

Volume pricing, team plans, and integration partnerships.

[email protected]

Contact form

Compose here, send from your mail app

Pick a topic and add the details. We'll open a draft in your mail app; you decide when to send it.

System status

See whether the API is having trouble right now.

View status

Developer docs

Endpoints, model IDs, error contracts, and integration examples.

Open docs

What TokenLab does

One API key: Call chat, image, video, audio and embedding models with one key. Every chat model accepts the OpenAI Chat Completions format; other request formats are listed on the model's page.
Prices you can check: Each model page shows the maker's official price next to the TokenLab price, in the unit the model is billed in. Price changes are recorded in the changelog.
Pay as you go: No subscription and no seat licence. You add funds, and each request is deducted at the listed rate. The balance does not expire.
A place to try models: The Console runs a prompt against a model before you write code, keeps the result, and shows what a failed request needs changed.

Where the numbers on this site come from

Model pages, comparisons and rankings mix several kinds of information. Each has its own source.

Prices: Official prices come from each maker's published price list. TokenLab prices are the rates charged on this service. Both are shown with the unit they apply to.
Capabilities and limits: Context window, output limit, supported inputs and operations come from the model catalog that the API itself uses, so a page and a request agree.
Success rate and latency: Measured on requests that went through TokenLab: success rate over the last 30 days, median and 95th-percentile latency over the last 7 days. A figure is published only when enough requests from enough separate accounts back it. Until then the page says it is still collecting data.
Rankings: Ordered by the request volume observed on TokenLab in the last 30 days. They show what customers of this service use; they are not a quality benchmark.

How model pages are written

The description, strengths, limits, use cases and questions on a model page are written for that model from its maker's own material: model cards, API references and release notes. The sources and the review date are listed at the bottom of each page.

A benchmark result appears only with the benchmark named and its publisher linked. What could not be verified is left out. Prices and limits are not part of the written text; the page renders them from current data.

Found something wrong? Write to [email protected] with the page address and we will correct it.