Divergent Labs
Two things. Small models trained for one job inside a product, which your code calls by the token. And GPUs by the minute to train and measure models of your own, in private beta.
Models
Specialized models for product teams: each one is trained to do one job inside a product well, and nothing else. Your code sends the thing — a chat, an email, a ticket — as JSON, and gets one checked line back. Billed by the token, from the org’s credit.
| Model | What it does | Price |
|---|---|---|
| Rubric | a chat’s title, and a one-line preview of an email, message, ticket or agent action | $0.10 per million input tokens, output free |
How models are called and billed →
Compute (private beta)
GPUs by the minute to train and measure models of your own, driven from the div CLI or, by an agent, from the MCP server. A run is placed, metered, streamed and stopped at a spend ceiling you set in dollars.
div login
div run --gpu h100 train.py::run --max-spend 5.00Compute is open to the orgs we have let in, and its commands, flags and prices can change before it opens to everyone. Request access.
Every page is Markdown too
Each page on this site is also served as the Markdown it was written in: add .md to the address, or use Copy page at the top of it. /llms.txt lists them all, for an agent that wants to read the docs before it calls a model or drives the CLI.