DEEPSYNTH
Agents search multiple websites and combine evidence to answer 120 research tasks across seven domains. They return structured JSON answers scored with precision, recall, F1, and exact match. A separate model judge assesses answers beyond exact matching.
The 40 development tasks include gold answers and decompositions. The 80 test tasks withhold answers for official evaluation. The current page lists twelve baseline systems, including repeated base models under different agent frameworks.
- Made by
- Huawei Noah's Ark Lab and university collaborators
- Released
- February 2026
- Kind
- Deep research
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- Public, plus a private test set
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- Browser
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- A host
- In views
- Agents
- Strongest model tested
- GPT-5.2 Pro #32 on the capability index
- Last model added
- December 2025
- Scores refreshed
- November 18th, 2026
In the authors’ words
Large language model (LLM)-based agents are increasingly used to solve complex tasks involving tool use, such as web browsing, code execution, and data analysis. However, current evaluation benchmarks do not adequately assess their ability to solve real-world tasks that require synthesizing information from multiple sources and inferring insights beyond simple fact retrieval. To address this, we introduce DEEPSYNTH, a novel benchmark designed to evaluate agents on realistic, time-consuming problems that combine information gathering, synthesis, and structured reasoning to produce insights. DEEPSYNTH contains 120 tasks collected across 7 domains and data sources covering 67 countries. DEEPSYNTH is constructed using a multi-stage data collection pipeline that requires annotators to collect official data sources, create hypotheses, perform manual analysis, and design tasks with verifiable answers. When evaluated on DEEPSYNTH, 11 state-of-the-art LLMs and deep research agents achieve a maximum F1 score of 8.97 and 17.5 on the LLM-judge metric, underscoring the difficulty of the benchmark. Our analysis reveals that current agents struggle with hallucinations and reasoning over large information spaces, highlighting DEEPSYNTH as a crucial benchmark for guiding future research.
OpenAI deep research leads with 8.97% among models released by December 2025 OpenAI deep research leads with 8.97% among models released by December 2025
Best score per entry from the benchmark's own leaderboard, read on November 18th, 2026. Where a model ran on its own with several settings, we keep its best. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. Agents built around a model, such as Smolagent-GPT5 + GPT-5, are their own entries. We don't count them as the model's score. This view shows only models we track, each at its best.
The newest model with a score here came out in December 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1 OpenAI deep research test, Pass@1 8.97%
- 2 GPT-5.2 Pro test, Pass@1 8.70%
- 3 Smolagent-GPT5 + GPT-5 test, Pass@1 6.42%
- 4 gemini-pro-2.5 (2025-08) test, Pass@1 6.25%
- 5 OWL-GPT4.1 + GPT-4.1 test, Pass@1 5.41%
- 6 GPT-5.1 test, Pass@1 3.83%
- 7 Smolagent-GPT4.1 + GPT-4.1 test, Pass@1 3.75%
- 8 GPT-4.1 test, Pass@1 3.46%
- 9 o3 test, Pass@1 3.29%
- 10 deepseek-r1-chat (2025-08) test, Pass@1 3.23%
- 11 o4-mini test, Pass@1 3.05%
- 12 deepseek-r1 (2026-02) test, Pass@1 2.80%
Source: Official DEEPSYNTH leaderboard
- 1 OpenAI deep research test, Pass@1 8.97%
- 2 GPT-5.2 Pro test, Pass@1 8.70%
- 3 GPT-5.1 test, Pass@1 3.83%
- 4 GPT-4.1 test, Pass@1 3.46%
- 5 o3 test, Pass@1 3.29%
- 6 o4-mini test, Pass@1 3.05%
Source: Official DEEPSYNTH leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | OpenAI deep research | 8.97% | test, Pass@1, o3-deep-research | Feb 2025 | Official leaderboard |
| 2 | GPT-5.2 Pro | 8.70% | test, Pass@1, GPT-5.2-Pro | Dec 2025 | Official leaderboard |
| 3 | Smolagent-GPT5 + GPT-5 | 6.42% | test, Pass@1 | Aug 2025 | Official leaderboard |
| 4 | gemini-pro-2.5 (2025-08) | 6.25% | test, Pass@1, Gemini-Pro-2.5 | Official leaderboard | |
| 5 | OWL-GPT4.1 + GPT-4.1 | 5.41% | test, Pass@1 | Apr 2025 | Official leaderboard |
| 6 | GPT-5.1 | 3.83% | test, Pass@1, GPT-5.1 | Nov 2025 | Official leaderboard |
| 7 | Smolagent-GPT4.1 + GPT-4.1 | 3.75% | test, Pass@1 | Apr 2025 | Official leaderboard |
| 8 | GPT-4.1 | 3.46% | test, Pass@1, GPT-4.1 | Apr 2025 | Official leaderboard |
| 9 | o3 | 3.29% | test, Pass@1, o3 | Apr 2025 | Official leaderboard |
| 10 | deepseek-r1-chat (2025-08) | 3.23% | test, Pass@1, DeepSeek-R1-Chat | Official leaderboard | |
| 11 | o4-mini | 3.05% | test, Pass@1, o4-mini | Apr 2025 | Official leaderboard |
| 12 | deepseek-r1 (2026-02) | 2.80% | test, Pass@1, DeepSeek-R1-Reasoner | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | OpenAI deep research | 8.97% | test, Pass@1, o3-deep-research | Feb 2025 | Official leaderboard |
| 2 | GPT-5.2 Pro | 8.70% | test, Pass@1, GPT-5.2-Pro | Dec 2025 | Official leaderboard |
| 3 | GPT-5.1 | 3.83% | test, Pass@1, GPT-5.1 | Nov 2025 | Official leaderboard |
| 4 | GPT-4.1 | 3.46% | test, Pass@1, GPT-4.1 | Apr 2025 | Official leaderboard |
| 5 | o3 | 3.29% | test, Pass@1, o3 | Apr 2025 | Official leaderboard |
| 6 | o4-mini | 3.05% | test, Pass@1, o4-mini | Apr 2025 | Official leaderboard |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official DEEPSYNTH leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official DEEPSYNTH leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Tool use
- AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state.
Gemini 4 Neon 56.3%
- MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks.
Gemini 4 Argon 90.7%
- BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes.
Qwen3.7-Max 75%
- τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state.
Qwen 3.8 Max 55.1%
- Toolathlon-Verified In use 108 long-horizon tasks across 32 apps through about 600 MCP tools. Execution-based checkers inspect final environment state. GLM-5.3-Flash 78.4%
- DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 11 months ago, in December 2025, for GPT-5.2 Pro. Every benchmark
Catalog record 2 pass
- Pass Task access Public tasks, plus a private test set.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the official leaderboard (12 entries).
- Pass Models scored 6 models we track, plus 6 other entries on its leaderboard.
- Warning Newest score The newest came 11 months ago, in December 2025, for GPT-5.2 Pro.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.2 Pro (#32).
- Pass Headroom The best is 8.97%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on huggingface.co: 9 rows for models we track and 3 other entries at the last read.
- Pass Last read November 18th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.