Study finds coding agent harnesses increase costs without improving performance
Key point
A study shows that switching between harnesses like Claude Code, OpenCode, or mini-SWE-agent barely improves pass rates, while costs per task can triple due to system prompts being resent.
A controlled study has revealed that switching between coding-agent harnesses, such as Claude Code, OpenCode, or mini-SWE-agent, barely moves pass rates more than rerunning the same setup.
However, the cost per task can triple, driven mainly by the system prompt that gets resent with every step.
This item is a summary. The full, checkable version is at the original source:
Read the original at The Decoder ↗