AARRI-Bench (Claude Code) - Hands-on: leaderboard

Metric: Task success rate (%) under the 0/1 reward on the hands-on tasks: coding, experimental setup and data processing; 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 February 2028. 6 models tracked.

Top models

#ModelScore
1Claude Opus 4.757.1
2Qwen 3.6 Plus50
3MiniMax-M2.750
4GPT-5.3 Codex50
5Claude Sonnet 4.635.7
6Kimi K2.635.7

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