Evaluation

Round by round

Coding performance on multi-step tasksSPIRITT

SPIRITT's harness came out ahead end to end.

Model choiceSPIRITT

SPIRITT is not tied to one lab.

Ship and operateSPIRITT

Deploys, hosts and maintains the result.

Native developer surfacesCodex

Codex lives in the CLI, IDE and ChatGPT.

Pick SPIRITT if

You want top coding plus everything after the code.

Pick Codex if

You want OpenAI's coding agent inside your existing tools.