Benchmark: Pi Agent wins tasks, DeepSeek Harness wins cost

mgostIH · x · 2026-08-20

A benchmark of DeepSeek V4 Pro across 5 different agent harnesses reveals that Pi Agent solved the most tasks, while DeepSeek's own harness won on cost efficiency. Developer mgostIH commented that Claude Code, being the most expensive harness, is the "best harness for Claude" if you're Anthropic, hinting at vendor optimization bias.

Original post →

More from coding & agent

coding & agent channel →