Complete multimodal multi-hop research trajectories combining image analysis, tool calling, and embedded image understanding with text-based search. Built for MM-BrowseComp and OmniGAIA.
Available Now · Sample Available
Multimodal Deep Research
Multimodal research trajectories with image analysis and tool calling.
Volume
0
Volume
Scalable Volume
Coverage
Research
Model Benchmark
0.0
%
Claude Opus 4.6 · Pass at k
Targets
N/A
Modality & Delivery
ImageText→Text
Formats
JSONL (.jsonl)Images
What You Receive
Coverage
Package Structure
Differentiators
- Combines image analysis and embedded-image understanding with text search and tool calling.
Record Schema
| Field | Description |
|---|---|
final_report | The reference research deliverable produced from the trajectory. |
image_refs | Paths to the figures, charts, or documents the task depends on. |
rubric | The normalized scoring rubric used to grade reports. |
task | The multimodal research question or brief assigned to the agent. |
trajectory | The expert research trajectory: plans, searches, source evaluations, and resolutions. |
Results
| Subject | Type | Metric | Result |
|---|---|---|---|
| Claude Opus 4.6 | Pass@5 | ||
| 0% Pass@5 on both supplied tasks. | |||
| Reference trajectories | Trajectory reward | Mean outcome reward | |
| Mean outcome reward across the two successful reference trajectories. | |||
Methodology & Quality
Methodology
SME involvement
Domain-expert curated with up to L2 review
Stressors
Cross-modal evidence gathering and tool-mediated research.
Design criteria
- Capture complete multimodal multi-hop research trajectories.
Build assumptions
Quality approach
Known limitations
Customization & Scalability
Custom sets available. Additional volume in progress