Claude Code Task Costs Are 9x Higher Than Kimi Code in Composio Benchmark

Composio benchmark reveals Claude Code costs $2 per task versus $0.22 for Kimi Code, with token usage varying up to 30x across frameworks.

Woofun AI data shows that AI Agent infrastructure provider Composio integrated Kimi K3 with Kimi Code, Hermes, and Claude Code to execute 28 identical tasks. While success rates differed by only two tasks, median token usage ranged from 61,000 for Kimi Code to 340,000 for Claude Code. Estimated average costs were $0.22, $0.28, and $2 respectively, with individual task token consumption varying by up to 30 times. Hermes achieved the fastest median completion time of 179 seconds, compared to 297 seconds for Kimi Code and 348 seconds for Claude Code.

A separate study by Writer corroborated these findings across 22 enterprise tasks using six models. Replacing only the execution framework resulted in a 38% reduction in token usage and a 41% decrease in per-task costs. Completion time dropped by 44% while maintaining consistent quality levels.