DRNoise: leaderboard
DRNoise (2026) evaluates deep-research agents on 100 answer-recovery tasks whose gold answer must be assembled from two indirect record chains, all run in the same search-only BrowseComp-Plus agent. This board is accuracy in percent after one plausible document that states a conflicting answer is seeded into the searchable corpus; higher is better.
Metric: Noisy-condition accuracy (%; 100 deep-research tasks whose unique gold answer must be recovered from two corroborating indirect record chains, run in a fixed BrowseComp-Plus search-only ReAct agent (top-5 dense retrieval, 512-token snippets, up to 100 iterations); mean of three runs, with one direct false summary document added per task; higher is better). Source: arxiv.org. Saturation forecast: Around May 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.5 Flash | 26 | #101 |
| 2 | GPT-5.4 | 15.7 | #123 |
| 3 | Qwen 3 32B | 5.7 | #960 |
| 4 | DeepSeek V4 Flash | 1 | #258 |
| 5 | MiMo-V2.5 | 1 | #354 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=drnoise · How It Works · Data refreshed daily, snapshot 2026-11-13.