MV-Bench Multi-View Interfaces: leaderboard
MV-Bench (Shandong University, 2026) asks multimodal models, queried once as coding agents, to build a coordinated multi-view visualization interface in React, TypeScript and D3 from a reference screenshot, dataset and interaction specification derived from Tableau workbooks. The score is 0 to 100: executability times the mean of visual fidelity, data-binding correctness and interaction completeness, averaged over interfaces; higher is better.
Metric: Overall score (0-100): executability gate times the mean of static visual fidelity, data-binding correctness and interaction completeness, averaged per interface over the full-benchmark evaluation subset (72 base interfaces and 31 derived samples); each model is invoked as a coding agent through the Claude Agent SDK to build a React, TypeScript and D3 coordinated multi-view interface from a screenshot, dataset and interaction specification; single pass from one fixed prompt, no execution feedback; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 30.9 | #256 |
| 2 | Qwen 3.5 Plus | 25.51 | #232 |
| 3 | Claude Sonnet 4.5 | 25.27 | #250 |
| 4 | GPT-5.4 | 25.25 | #123 |
| 5 | GLM-4.6V | 5.92 | #701 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mv-bench-multi-view-interfaces · How It Works · Data refreshed daily, snapshot 2026-11-13.