MP-Bench (Failure Attribution) - Hand-Crafted MAS: leaderboard
Metric: nDCG@5 with exponential gain (0-1, times 100) of the model's ranking of failure-inducing steps against the ranking by annotator consensus, on MP-Bench's 169 failed hand-crafted (Magentic-One) multi-agent executions from GAIA and AssistantBench, three expert annotators per execution log; the model runs All-at-Once attribution zero-shot 3 times at temperature 1.0 and its runs are consolidated by consensus rate; higher is better. Source: arxiv.org. Saturation forecast: Around August 2028. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 43.97 | #138 |
| 2 | O3 Mini | 43.67 | #266 |
| 3 | GPT-4.1 | 43.13 | #240 |
| 4 | GPT-OSS-120B | 42.45 | #330 |
| 5 | GPT-5.1 | 37.47 | #131 |
| 6 | Qwen 3 8B | 29.44 | #667 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=mp-bench-failure-attribution-hand-crafted-mas · How It Works · Data refreshed daily, snapshot 2026-10-11.