Command Code Ties for Top Spot in 29-Task AI Coding Harness Benchmark

Key Info

Command Code, an agentic coding harness, reportedly matches Codex and Hermes Agent as the highest-performing harnesses in a 29-task benchmark comparison, finishing 21/29 while claiming top-tier token efficiency.

Highlights

• Tied for first among six harnesses with a 21/29 task success rate, alongside Codex and Hermes Agent; Claude Code scored 20/29, while OpenCode and Pi Agent scored 19/29.
• The spread was only about 7 percentage points: just 3 of 29 tasks produced different results across harnesses, with 18 tasks passed by all and 8 tasks failed by all.