TaskEvolving Deep Research (Agent, With Hints): leaderboard
TaskEvolving (Wang, Chen et al., 2026) evolves 500 NQ-Open questions into verifiable deep research tasks, each a graph of atomic steps with fact-grounded rubric checkpoints scored by a Qwen3.7-Max judge. In this setting the model searches the web with Google search and page scraping tools and receives the full query with its analysis dimensions and constraints. Score is the percentage of rubric points met; higher is better.
Metric: Rubric score (%; fact-grounded pointwise checkpoints of 500 evolved deep research tasks, evidential plus analytical, judged by Qwen3.7-Max; agent mode with Google search and web scraping tools; query with hints: analysis dimensions and solving constraints stated). Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Terra | 86 | #60 |
| 2 | DeepSeek V4 Pro | 83 | #181 |
| 3 | Qwen 3.7 Max | 83 | #142 |
| 4 | GLM-5.2 | 80 | #137 |
| 5 | Claude Sonnet 5 | 80 | #108 |
| 6 | Gemini 3.5 Flash | 74 | #98 |
| 7 | MiniMax-M2.5 | 67 | #590 |
| 8 | Qwen 3.6 Flash | 53 | #422 |
| 9 | Qwen 3 32B | 36 | #960 |
| 10 | Qwen 3 14B | 35 | #1235 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=taskevolving-deep-research-agent-with-hints · How It Works · Data refreshed daily, snapshot 2026-11-12.