VDR-Bench
Models answer 2,000 visual questions that require finding entities in images and researching related facts. A Qwen3-VL judge checks final answers against references. Entity recall separately measures whether searches retrieve the relevant entities.
The paper also proposes a cropped-image search workflow, but releases a separate benchmark and compares multiple research systems. The shared repository star count is not benchmark-specific. Results depend on search-engine behavior and judge decisions.
- Released
- February 2026
- Kind
- Deep research
- Level
- University
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- Browser
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- GPT-5 #71 on the capability index
- Last model added
- October 2025
- Scores refreshed
- November 18th, 2026
In the authors’ words
Multimodal Large Language Models (MLLMs) have advanced VQA and now support Vision-DeepResearch systems that use search engines for complex visual-textual fact-finding. However, evaluating these visual and textual search abilities is still difficult, and existing benchmarks have two major limitations. First, existing benchmarks are not visual search-centric: answers that should require visual search are often leaked through cross-textual cues in the text questions or can be inferred from the prior world knowledge in current MLLMs. Second, overly idealized evaluation scenario: On the image-search side, the required information can often be obtained via near-exact matching against the full image, while the text-search side is overly direct and insufficiently challenging. To address these issues, we construct the Vision-DeepResearch benchmark (VDR-Bench) comprising 2,000 VQA instances. All questions are created via a careful, multi-stage curation pipeline and rigorous expert review, designed to assess the behavior of Vision-DeepResearch systems under realistic real-world conditions. Moreover, to address the insufficient visual retrieval capabilities of current MLLMs, we propose a simple multi-round cropped-search workflow. This strategy is shown to effectively improve model performance in realistic visual retrieval scenarios. Overall, our results provide practical guidance for the design of future multimodal deep-research systems. The code will be released in https://github.com/Osilly/Vision-DeepResearch.
Vision-DeepResearch-30B-A3B tops the leaderboard with 37.8% Claude 3.7 Sonnet leads with 27.2% among models released by October 2025
Best score per entry from the benchmark's own leaderboard, read on November 18th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Claude 3.7 Sonnet, at 27.2%. This view shows only models we track, each at its best.
The newest model with a score here came out in October 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 Vision-DeepResearch-30B-A3B agentic search 37.8%
- 2 Vision-DeepResearch-8B agentic search 29.2%
- 3 Claude 3.7 Sonnet agentic search 27.2%
- 4
Qwen3-VL-30B-A3B-Thinking agentic search 23.2%
- 5 GPT-5 agentic search 20.4%
- 6
Qwen3-VL-30B-A3B-Instruct agentic search 20.2%
- 7 Gemini-2.5 Pro (snapshot unspecified) agentic search 18.8%
- 8
Qwen3-VL-8B-Thinking agentic search 17.6%
- 9
Qwen3-VL-8B-Instruct agentic search 17%
- 10
Gemini 2.5 Flash Sep 2025, agentic search 16.3%
- 11 Claude Sonnet 4 agentic search 13.6%
Source: Official VDR-Bench leaderboard
- 1 Claude 3.7 Sonnet agentic search 27.2%
- 2
Qwen3-VL-30B-A3B-Thinking agentic search 23.2%
- 3 GPT-5 agentic search 20.4%
- 4
Qwen3-VL-30B-A3B-Instruct agentic search 20.2%
- 5
Qwen3-VL-8B-Thinking agentic search 17.6%
- 6
Qwen3-VL-8B-Instruct agentic search 17%
- 7
Gemini 2.5 Flash Sep 2025, agentic search 16.3%
- 8 Claude Sonnet 4 agentic search 13.6%
Source: Official VDR-Bench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Vision-DeepResearch-30B-A3B | 37.8% | agentic search, Vision-DeepResearch | Official leaderboard | |
| 2 | Vision-DeepResearch-8B | 29.2% | agentic search, Vision-DeepResearch | Official leaderboard | |
| 3 | Claude 3.7 Sonnet | 27.2% | agentic search, Vision-DeepResearch | Feb 2025 | Official leaderboard |
| 4 | | 23.2% | agentic search, Vision-DeepResearch | Oct 2025 | Official leaderboard |
| 5 | GPT-5 | 20.4% | agentic search, Vision-DeepResearch | Aug 2025 | Official leaderboard |
| 6 | | 20.2% | agentic search, Vision-DeepResearch | Oct 2025 | Official leaderboard |
| 7 | Gemini-2.5 Pro (snapshot unspecified) | 18.8% | agentic search, Vision-DeepResearch | Official leaderboard | |
| 8 | | 17.6% | agentic search, Vision-DeepResearch | Oct 2025 | Official leaderboard |
| 9 | | 17% | agentic search, Vision-DeepResearch | Oct 2025 | Official leaderboard |
| 10 | | 16.3% | agentic search, Vision-DeepResearch | Sep 2025 | Official leaderboard |
| 11 | Claude Sonnet 4 | 13.6% | agentic search, Vision-DeepResearch | May 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude 3.7 Sonnet | 27.2% | agentic search, Vision-DeepResearch | Feb 2025 | Official leaderboard |
| 2 | | 23.2% | agentic search, Vision-DeepResearch | Oct 2025 | Official leaderboard |
| 3 | GPT-5 | 20.4% | agentic search, Vision-DeepResearch | Aug 2025 | Official leaderboard |
| 4 | | 20.2% | agentic search, Vision-DeepResearch | Oct 2025 | Official leaderboard |
| 5 | | 17.6% | agentic search, Vision-DeepResearch | Oct 2025 | Official leaderboard |
| 6 | | 17% | agentic search, Vision-DeepResearch | Oct 2025 | Official leaderboard |
| 7 | | 16.3% | agentic search, Vision-DeepResearch | Sep 2025 | Official leaderboard |
| 8 | Claude Sonnet 4 | 13.6% | agentic search, Vision-DeepResearch | May 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 VDR-Bench 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 VDR-Bench 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 13 months ago, in October 2025, for Qwen3-VL-8B-Instruct. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the official leaderboard (11 entries).
- Pass Models scored 8 models we track, plus 3 other entries on its leaderboard.
- Warning Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-8B-Instruct.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5 (#71).
- Pass Headroom The best is 27.2%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on raw.githubusercontent.com: 16 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.