DeepResearch Bench
ReportedNo description yet: we only describe an eval set when we can do it correctly.
- Kind
- Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
- Link
- https://evals.futuresearch.ai/#drb
- Score shown
- Epoch's column "Average score", as a percentage, higher is better; random guessing scores about 0.0%. The source gives no error bars.
- Results
- 41 in all: 1 for open-weight models, 40 for closed ones. 1 linked to a model page here.
Best open-weight result: deepseek-ai/DeepSeek-R1-0528, 35.1%, number 40 of 41 results.
| # | Model | Weights | Setting | Score | Date | Source |
|---|---|---|---|---|---|---|
| 1 | claude-opus-4-6_highAnthropic | closed | effort high | 55.3% | not dated | evals.futuresearch.ai |
| 2 | claude-sonnet-4-6_highAnthropic | closed | effort high | 54.9% | not dated | evals.futuresearch.ai |
| 3 | claude-opus-4-5-20251101_highAnthropic | closed | effort high | 54.8% | not dated | evals.futuresearch.ai |
| 4 | gpt-5.5_highOpenAI | closed | effort high | 54.0% | not dated | evals.futuresearch.ai |
| 5 | claude-opus-4-5-20251101_lowAnthropic | closed | effort low | 53.7% | not dated | evals.futuresearch.ai |
| 6 | claude-opus-4-6_mediumAnthropic | closed | effort medium | 53.2% | not dated | evals.futuresearch.ai |
| 7 | claude-sonnet-4-5-20250929_2KAnthropic | closed | default | 52.6% | not dated | DeepResearchBench Leaderboard |
| 8 | claude-opus-4-6_lowAnthropic | closed | effort low | 51.4% | not dated | evals.futuresearch.ai |
| 9 | claude-sonnet-4-6_lowAnthropic | closed | effort low | 50.4% | not dated | evals.futuresearch.ai |
| 10 | claude-opus-4-8_highAnthropic | closed | effort high | 50.2% | not dated | evals.futuresearch.ai |
| 11 | gemini-3-flash-preview_lowGoogle DeepMind | closed | effort low | 49.8% | not dated | evals.futuresearch.ai |
| 12 | gpt-5-2025-08-07_lowOpenAI | closed | effort low | 49.6% | not dated | evals.futuresearch.ai |
| 13 | gpt-5.5_mediumOpenAI | closed | effort medium | 49.6% | not dated | evals.futuresearch.ai |
| 14 | claude-opus-4-8_lowAnthropic | closed | effort low | 49.3% | not dated | evals.futuresearch.ai |
| 15 | gemini-3-flash-preview_minimalGoogle DeepMind | closed | effort minimal | 49.0% | not dated | evals.futuresearch.ai |
| 16 | gpt-5-2025-08-07_minimalOpenAI | closed | effort minimal | 48.9% | not dated | evals.futuresearch.ai |
| 17 | gpt-5.5_lowOpenAI | closed | effort low | 48.7% | not dated | evals.futuresearch.ai |
| 18 | gpt-5-2025-08-07_mediumOpenAI | closed | effort medium | 48.6% | not dated | evals.futuresearch.ai |
| 19 | claude-opus-4-1-20250805Anthropic | closed | default | 48.3% | not dated | evals.futuresearch.ai |
| 20 | gpt-5-2025-08-07_highOpenAI | closed | effort high | 48.1% | not dated | evals.futuresearch.ai |
| 21 | gemini-3-flash-preview_highGoogle DeepMind | closed | effort high | 47.9% | not dated | evals.futuresearch.ai |
| 22 | gemini-3.1-pro-preview_highGoogle DeepMind | closed | effort high | 47.8% | not dated | evals.futuresearch.ai |
| 23 | claude-sonnet-4-5-20250929_lowAnthropic | closed | effort low | 47.5% | not dated | evals.futuresearch.ai |
| 24 | claude-opus-4-8_mediumAnthropic | closed | effort medium | 47.4% | not dated | evals.futuresearch.ai |
| 25 | grok-4-0709xAI | closed | Agent: True | 47.3% | not dated | evals.futuresearch.ai |
| 26 | claude-opus-4-20250514_2KAnthropic | closed | default | 46.8% | not dated | evals.futuresearch.ai |
| 27 | claude-sonnet-4-20250514_2KAnthropic | closed | default | 46.6% | not dated | evals.futuresearch.ai |
| 28 | gemini-3-pro-preview_lowGoogle DeepMind | closed | effort low | 46.3% | not dated | evals.futuresearch.ai |
| 29 | claude-haiku-4-5-20251001_lowAnthropic | closed | effort low | 45.5% | not dated | evals.futuresearch.ai |
| 30 | o3-2025-04-16_mediumOpenAI | closed | effort medium | 45.2% | not dated | evals.futuresearch.ai |
| 31 | gemini-3.1-pro-preview_lowGoogle DeepMind | closed | effort low | 44.5% | not dated | evals.futuresearch.ai |
| 32 | claude-3-7-sonnet-20250219_2KAnthropic | closed | default | 43.6% | not dated | DeepResearchBench Leaderboard |
| 33 | gemini-2.5-pro-preview-06-05Google DeepMind | closed | default | 42.8% | not dated | DeepResearchBench Leaderboard |
| 34 | gpt-5.1-2025-11-13_lowOpenAI | closed | effort low | 42.8% | not dated | evals.futuresearch.ai |
| 35 | gemini-2.5-proGoogle DeepMind | closed | default | 41.5% | not dated | evals.futuresearch.ai |
| 36 | gpt-5.2-2025-12-11_lowOpenAI | closed | effort low | 41.1% | not dated | evals.futuresearch.ai |
| 37 | gemini-3.1-flash-lite_lowGoogle | closed | effort low | 37.3% | not dated | evals.futuresearch.ai |
| 38 | gemini-3.1-flash-liteGoogle | closed | default | 36.4% | not dated | evals.futuresearch.ai |
| 39 | gpt-5.4-mini-2026-03-17_lowOpenAI | closed | effort low | 36.3% | not dated | evals.futuresearch.ai |
| 40 | deepseek-ai/DeepSeek-R1-0528Epoch: DeepSeek-R1-0528, DeepSeek | open | default | 35.1% | not dated | DeepResearchBench Leaderboard |
| 41 | gpt-5.4-2026-03-05_lowOpenAI | closed | effort low | 35.1% | not dated | evals.futuresearch.ai |
Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method