AARRI-Bench (Claude Code): leaderboard
Metric: Task success rate (%) under the 0/1 reward (a task counts only when every one of its unit tests passes) over the 82 hand-crafted AI-research-intern tasks in four scenario categories; 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 November 2027. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.7 | 62.2 |
| 2 | Qwen 3.6 Plus | 56.3 |
| 3 | MiniMax-M2.7 | 56.1 |
| 4 | GPT-5.3 Codex | 53.1 |
| 5 | Claude Sonnet 4.6 | 52.4 |
| 6 | Kimi K2.6 | 51.3 |
Interactive version: theaggregate.ai/benchmark?slug=aarri-bench-claude-code · How It Works · Data refreshed daily, snapshot 2026-09-29.