AARRI-Bench (Claude Code) - Interaction: leaderboard

Metric: Task success rate (%) under the 0/1 reward on the interaction tasks: using research tools and collaborating with human stakeholders over multi-turn exchanges; Harbor containerized runs, Claude Code agent harness; a configuration's tasks without a valid run are left out of its denominator; higher is better. Source: arxiv.org. Saturation forecast: Around August 2027. 6 models tracked.

Top models

#ModelScore
1Claude Opus 4.766.7
2MiniMax-M2.766.7
3Kimi K2.665
4GPT-5.3 Codex65
5Qwen 3.6 Plus63.2
6Claude Sonnet 4.661.9

Interactive version: theaggregate.ai/benchmark?slug=aarri-bench-claude-code-interaction · How It Works · Data refreshed daily, snapshot 2026-09-29.