Composio Tests Kimi K3 on Six Agent Frameworks: Claude Code Is the Most Expensive and Slowest

According to Beating Monitor, AI agent infrastructure company Composio expanded its testing by integrating the same Kimi K3 model into six different agent frameworks, completing 26 identical tasks. The comparison focuses on the execution frameworks, with Kimi K3 as the underlying model. Kimi Code achieved first place in 21 tasks. Hermes completed 20, while Pi Agent and Claude Code each completed 19, OpenCode completed 18, and Codex ranked last with 17. Based on Kimi K3’s official pricing, Hermes and Pi Agent had the lowest average cost per task at $0.39 and $0.40, respectively. Claude Code reached $1.47—nearly 3.8 times higher than Hermes. Pi Agent was the fastest, with a median task completion time of 161.7 seconds, while Claude Code was the slowest at 347.6 seconds. Despite using the same model, the difference in successful task completion across frameworks was at most four tasks, yet average costs varied by nearly fourfold and completion times differed by more than twofold.