DeepSearchQA
Research questions with many separate facts to gather; scored on answers that are fully correct, so missing one item fails the task. Published by Google.
As of 21 Sept 2026, Gemini Deep Research Agent leads DeepSearchQA on BenchLeader with 66.1%, ahead of GPT-5 Pro at 65.2%, across 13 model configurations with a published result.
- Published by
- Googledata via Kaggle Benchmarks
- Category
- Agents & tools
- Index weight
- Reference only
- Models
- 13
- Data as of
- 21 Sept 2026
Kaggle (kaggle.com); each board credits its publishing lab.
- 1Gemini Deep Research Agent66.1%
- 2GPT-5 Pro65.2%
- 3GPT-5.463.7%
- 4Gemini 3.1 Pro60.2%
- 5GPT-559.4%
- 6Claude Sonnet 4.650.8%
- 7o3 Deep Research44.2%
- 8o4 mini Deep Research40.4%
- 9Gemini 2.5 Flash25.9%
- 10Gemini 2.5 Pro24.6%
- 11Claude Opus 4.5 (thinking)24.0%
- 12Claude Sonnet 4.5 (thinking)16.0%
- 13Claude Haiku 4.5 (thinking)12.8%
13 of 13
| # | ||||
|---|---|---|---|---|
| 1 |  Gemini Deep Research AgentGoogle | 66.1% | – | 2025-12-11 |
| 2 |  GPT-5 ProOpenAI | 65.2% | 60.5 | 2025-12-11 |
| 3 | GPT-5.4OpenAI | 63.7% | 59.3 | – |
| 4 | Gemini 3.1 ProGoogle | 60.2% | 63.8 | 2026-04-13 |
| 5 | GPT-5OpenAI | 59.4% | 57.2 | 2025-12-11 |
| 6 |  Claude Sonnet 4.6Anthropic | 50.8% | 59.0 | 2026-04-13 |
| 7 | o3 Deep ResearchOpenAI | 44.2% | – | 2025-12-11 |
| 8 | o4 mini Deep ResearchOpenAI | 40.4% | – | 2025-12-11 |
| 9 | Gemini 2.5 FlashGoogle | 25.9% | 52.5 | 2025-12-11 |
| 10 | Gemini 2.5 ProGoogle | 24.6% | 54.4 | 2025-12-11 |
| 11 | Claude Opus 4.5thinkingAnthropic | 24.0% | 60.8 | 2025-12-11 |
| 12 | Claude Sonnet 4.5thinkingAnthropic | 16.0% | 53.3 | 2025-12-11 |
| 13 | Claude Haiku 4.5thinkingAnthropic | 12.8% | 47.4 | 2025-12-11 |
Cite as: BenchLeader, “DeepSearchQA leaderboard”, https://www.benchleader.com/benchmarks/kaggle_dsqa, data as of 21 Sept 2026.
DeepSearchQA: questions
- What does DeepSearchQA measure?
- Research questions with many separate facts to gather; scored on answers that are fully correct, so missing one item fails the task. Published by Google. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads DeepSearchQA?
- Gemini Deep Research Agent leads DeepSearchQA with 66.1% as of 21 Sept 2026, ahead of GPT-5 Pro at 65.2%.
- How many models have DeepSearchQA results?
- 13 model configurations have a DeepSearchQA result on BenchLeader, all taken from Google via Kaggle Benchmarks.
- Who runs DeepSearchQA and how often is it updated?
- DeepSearchQA is published by Google. BenchLeader re-reads the published results every morning and records the date each result was published.
- Does DeepSearchQA count toward the BenchLeader Index?
- No. DeepSearchQA is shown for reference but left out of the composite index.