Skip to content
Weights Not Walls
Menu

DeepResearch Bench

Reported

No 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.

DeepResearch Bench results
#ModelWeightsSettingScoreDateSource
1claude-opus-4-6_highAnthropicclosedeffort high55.3%not datedevals.futuresearch.ai
2claude-sonnet-4-6_highAnthropicclosedeffort high54.9%not datedevals.futuresearch.ai
3claude-opus-4-5-20251101_highAnthropicclosedeffort high54.8%not datedevals.futuresearch.ai
4gpt-5.5_highOpenAIclosedeffort high54.0%not datedevals.futuresearch.ai
5claude-opus-4-5-20251101_lowAnthropicclosedeffort low53.7%not datedevals.futuresearch.ai
6claude-opus-4-6_mediumAnthropicclosedeffort medium53.2%not datedevals.futuresearch.ai
7claude-sonnet-4-5-20250929_2KAnthropiccloseddefault52.6%not datedDeepResearchBench Leaderboard
8claude-opus-4-6_lowAnthropicclosedeffort low51.4%not datedevals.futuresearch.ai
9claude-sonnet-4-6_lowAnthropicclosedeffort low50.4%not datedevals.futuresearch.ai
10claude-opus-4-8_highAnthropicclosedeffort high50.2%not datedevals.futuresearch.ai
11gemini-3-flash-preview_lowGoogle DeepMindclosedeffort low49.8%not datedevals.futuresearch.ai
12gpt-5-2025-08-07_lowOpenAIclosedeffort low49.6%not datedevals.futuresearch.ai
13gpt-5.5_mediumOpenAIclosedeffort medium49.6%not datedevals.futuresearch.ai
14claude-opus-4-8_lowAnthropicclosedeffort low49.3%not datedevals.futuresearch.ai
15gemini-3-flash-preview_minimalGoogle DeepMindclosedeffort minimal49.0%not datedevals.futuresearch.ai
16gpt-5-2025-08-07_minimalOpenAIclosedeffort minimal48.9%not datedevals.futuresearch.ai
17gpt-5.5_lowOpenAIclosedeffort low48.7%not datedevals.futuresearch.ai
18gpt-5-2025-08-07_mediumOpenAIclosedeffort medium48.6%not datedevals.futuresearch.ai
19claude-opus-4-1-20250805Anthropiccloseddefault48.3%not datedevals.futuresearch.ai
20gpt-5-2025-08-07_highOpenAIclosedeffort high48.1%not datedevals.futuresearch.ai
21gemini-3-flash-preview_highGoogle DeepMindclosedeffort high47.9%not datedevals.futuresearch.ai
22gemini-3.1-pro-preview_highGoogle DeepMindclosedeffort high47.8%not datedevals.futuresearch.ai
23claude-sonnet-4-5-20250929_lowAnthropicclosedeffort low47.5%not datedevals.futuresearch.ai
24claude-opus-4-8_mediumAnthropicclosedeffort medium47.4%not datedevals.futuresearch.ai
25grok-4-0709xAIclosedAgent: True47.3%not datedevals.futuresearch.ai
26claude-opus-4-20250514_2KAnthropiccloseddefault46.8%not datedevals.futuresearch.ai
27claude-sonnet-4-20250514_2KAnthropiccloseddefault46.6%not datedevals.futuresearch.ai
28gemini-3-pro-preview_lowGoogle DeepMindclosedeffort low46.3%not datedevals.futuresearch.ai
29claude-haiku-4-5-20251001_lowAnthropicclosedeffort low45.5%not datedevals.futuresearch.ai
30o3-2025-04-16_mediumOpenAIclosedeffort medium45.2%not datedevals.futuresearch.ai
31gemini-3.1-pro-preview_lowGoogle DeepMindclosedeffort low44.5%not datedevals.futuresearch.ai
32claude-3-7-sonnet-20250219_2KAnthropiccloseddefault43.6%not datedDeepResearchBench Leaderboard
33gemini-2.5-pro-preview-06-05Google DeepMindcloseddefault42.8%not datedDeepResearchBench Leaderboard
34gpt-5.1-2025-11-13_lowOpenAIclosedeffort low42.8%not datedevals.futuresearch.ai
35gemini-2.5-proGoogle DeepMindcloseddefault41.5%not datedevals.futuresearch.ai
36gpt-5.2-2025-12-11_lowOpenAIclosedeffort low41.1%not datedevals.futuresearch.ai
37gemini-3.1-flash-lite_lowGoogleclosedeffort low37.3%not datedevals.futuresearch.ai
38gemini-3.1-flash-liteGooglecloseddefault36.4%not datedevals.futuresearch.ai
39gpt-5.4-mini-2026-03-17_lowOpenAIclosedeffort low36.3%not datedevals.futuresearch.ai
40deepseek-ai/DeepSeek-R1-0528Epoch: DeepSeek-R1-0528, DeepSeekopendefault35.1%not datedDeepResearchBench Leaderboard
41gpt-5.4-2026-03-05_lowOpenAIclosedeffort low35.1%not datedevals.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