What it is, in our words
A coding agent we run as the other home base, and the grader behind our published benchmark labels.
What we measured
Home base. It is a CLI, it is an MCP client, it can run as an MCP server, and the models behind it are callable through the API with an SDK to embed the agent.
We used Codex CLI 0.159.0 in a read-only sandbox at default reasoning settings to grade every candidate in the search API benchmark, and we saved every prompt verbatim.
Surface we measured: official MCP server, CLI available, full REST, tier S.
Vendor
Codex. We link out and make no claim about their current pricing, terms or roadmap.
