LH-Bench (Figma-to-Code): leaderboard
Metric: Output-tier artifact score (1-5): a Gemini 3 VLM judge compares Playwright screenshots of the built page with expert ground-truth images on eight weighted rubrics (visual fidelity, layout, interaction), mean over the tasks the configuration ran, on the 33 LH-Bench Figma-to-code tasks (rebuild a real Figma design as a working front end through the Figma API over MCP, following an expert-authored SKILL.md workflow), the model running autonomously in its vendor's agent harness; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 4.19 | #60 |
| 2 | GPT-5.2 | 3.94 | #105 |
| 3 | Claude Opus 4.5 | 3.88 | #79 |
| 4 | Gemini 3.1 Pro (Preview) | 3.73 | #54 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=lh-bench-figma-to-code · How It Works · Data refreshed daily, snapshot 2026-10-11.