Optimized models for agent work, built from real traces.
Customer service agent
built from τ²-benchMulti-turn customer-service tool use: look up a user, book or modify an order, and hold to the domain's policy across the conversation.
Quality, cost, and savings not yet measured
tool-callscustomer-servicemulti-turn
Terminal agent
built from Terminal-Bench 2.0Command-line work in a container: inspect a filesystem, run shell commands, and read their output to drive a task to completion.
Quality, cost, and savings not yet measured
terminalshellcontainers
Code repair agent
built from SWE-benchRepository-level bug fixing: read the failing test, navigate an unfamiliar codebase, and produce a patch that resolves the issue.
Quality, cost, and savings not yet measured
codepatchingrepositories
Build your own
+your traces → optimized model
Turn your own agent traces into a model served behind an OpenAI-compatible URL.