[Kimi K3 Integrated with Claude Code, Average Task Cost is 9 Times That of Kimi Code] According to monitoring by Beating, AI Agent infrastructure company Composio integrated the same Kimi K3 with Kimi Code, Hermes, and Claude Code to complete 28 identical tasks. The success rate across the three operational frameworks differed by only 2 tasks, but the token usage and costs varied significantly. Kimi Code completed 22 tasks, Hermes completed 21 tasks, and Claude Code completed 20 tasks. The median token usage per task was 61,000, 67,000, and 340,000, respectively. Composio estimated the average cost per task to be approximately $0.22, $0.28, and $2. For certain tasks, token usage differed by as much as 30 times. Hermes was the fastest, with a median task completion time of 179 seconds. Kimi Code took 297 seconds, and Claude Code took 348 seconds. Another study by Writer yielded similar results. Researchers replaced only the operational framework across 22 enterprise tasks and 6 models, resulting in a 38% reduction in token usage, a 41% reduction in per-task cost, a 44% reduction in time, while maintaining comparable completion quality. [Original Link]