Unreal Agent: Asynchronous tool calls (40% cheaper than Codex)

Unreal Agent is an open-source agent framework that does not block the model while making tool calls. This enables asynchronous execution of file/API calls, meaning that the agent won’t have to wait, allowing you to save up to 40% tokens compared to Codex and up to 20% compared to Pi without losing any quality on the Terminal-Bench 4.0 and DeepSWE 1.1 benchmarks.

Last week, we saw how using different frameworks results in equivalent quality, but can result in a factor of 2 difference in cost. Unreal Agent fixes this issue by making the code execute asynchronously, removing all the wasted tokens from idling. I highly suggest trying out their runner with your examples. No changes needed to your prompts!