Divergent Labs
Documentation for Divergent Labs models, GPU compute and sandboxes.
Divergent Labs has three products. All three draw on the same org credit.
Models
Small models that each do one job in a product, such as writing a chat title or an inbox preview. Your code sends the item as JSON and gets one line back. Billed per token.
| Model | What it does | Price |
|---|---|---|
| Rubric | Chat titles, and one-line previews of emails, messages, tickets and agent actions. | $0.10 per million input tokens. Output is free. |
Compute (private beta)
GPUs by the minute for training and evaluation jobs. Start a run from your code’s directory with the div CLI, or from an agent with the MCP server. Set a spend limit in dollars and the run stops when it reaches it. From $4.95 per H100 hour.
div login
div run --gpu h100 train.py::run --max-spend 5.00Compute is open to invited orgs, and its commands and prices can change before general availability. Request access.
Sandboxes
Linux machines for agents. Create one, run commands in it, stop it, and resume it later with its files and memory intact. Every sandbox runs Claude Code and OpenCode, so you can send it a task with div agent prompt. From 3¢ an hour, billed by the minute. Sandboxes have no GPU. For GPU jobs, use Compute.
SB=$(div sandbox create)
div sandbox exec $SB -- npm testDocs as Markdown
Every page is also available as Markdown: add .md to the URL, or use Copy page. /llms.txt lists every page, for agents that read the docs before calling the API or the CLI.