Skip to content

Benchmark

Deep Research Bench

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.

Models measured
25
Scores recorded
41
Sources
2
IQ weight
1.71

Score distribution

48%
Mean
48.1%
Median
8.4%
Std dev
37.3%
10th percentile
54%
90th percentile
35.1%
Min
35.1%90%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.71 for chat models, measured over 25 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Kimi K3Kimi Agent · max effort · internal· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.90%
2Claude Opus 4.6high· Epoch AI Benchmarking Hub data55.3%
3Claude Sonnet 4.6high· Epoch AI Benchmarking Hub data54.9%
4Claude Opus 4.5high· Epoch AI Benchmarking Hub data54.8%
5GPT 5.5high· Epoch AI Benchmarking Hub data54%
6Claude Opus 4.5low· Epoch AI Benchmarking Hub data53.7%
7Claude Opus 4.6medium· Epoch AI Benchmarking Hub data53.2%
8Claude Sonnet 4.52k· Epoch AI Benchmarking Hub data52.6%
9Claude Opus 4.6low· Epoch AI Benchmarking Hub data51.4%
10Claude Sonnet 4.6low· Epoch AI Benchmarking Hub data50.4%
11Claude Opus 4.8high· Epoch AI Benchmarking Hub data50.2%
12Gemini 3 Flashlow· Epoch AI Benchmarking Hub data49.8%
13GPT 5low· Epoch AI Benchmarking Hub data49.6%
14GPT 5.5medium· Epoch AI Benchmarking Hub data49.6%
15Claude Opus 4.8low· Epoch AI Benchmarking Hub data49.3%
16Gemini 3 Flashminimal· Epoch AI Benchmarking Hub data49%
17GPT 5minimal· Epoch AI Benchmarking Hub data48.9%
18GPT 5.5low· Epoch AI Benchmarking Hub data48.7%
19GPT 5medium· Epoch AI Benchmarking Hub data48.6%
20Claude Opus 4.1· Epoch AI Benchmarking Hub data48.3%
21GPT 5high· Epoch AI Benchmarking Hub data48.1%
22Gemini 3 Flashhigh· Epoch AI Benchmarking Hub data47.9%
23Gemini 3.1 Pro Previewhigh· Epoch AI Benchmarking Hub data47.8%
24Claude Sonnet 4.5low· Epoch AI Benchmarking Hub data47.5%
25Claude Opus 4.8medium· Epoch AI Benchmarking Hub data47.4%
26Grok 4· Epoch AI Benchmarking Hub data47.3%
27Claude Opus 42k· Epoch AI Benchmarking Hub data46.8%
28Claude Sonnet 42k· Epoch AI Benchmarking Hub data46.6%
29Gemini 3 Pro Previewlow· Epoch AI Benchmarking Hub data46.3%
30Claude Haiku 4.5low· Epoch AI Benchmarking Hub data45.5%
31OpenAI o3medium· Epoch AI Benchmarking Hub data45.2%
32Gemini 3.1 Pro Previewlow· Epoch AI Benchmarking Hub data44.5%
33Claude Sonnet 3.72k· Epoch AI Benchmarking Hub data43.6%
34Gemini 2.5 Pro· Epoch AI Benchmarking Hub data42.8%
35GPT 5.1low· Epoch AI Benchmarking Hub data42.8%
36GPT 5.2low· Epoch AI Benchmarking Hub data41.1%
37Gemini 3.1 Flash Litelow· Epoch AI Benchmarking Hub data37.3%
38Gemini 3.1 Flash Lite· Epoch AI Benchmarking Hub data36.4%
39GPT-5.4 minilow· Epoch AI Benchmarking Hub data36.3%
40DeepSeek R1· Epoch AI Benchmarking Hub data35.1%
41GPT 5.4low· Epoch AI Benchmarking Hub data35.1%

Data credits: Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.