MobileForge: New Benchmark for Multi-Screen Mobile App Generation
MobileForge has been launched as the inaugural benchmark aimed at project-level multi-screen mobile app development. This new benchmark overcomes the shortcomings of current design-to-code benchmarks, which primarily concentrate on single-page outputs and do not assess cross-page navigation or overall project maintainability. MobileForge includes actual mobile applications, screens reviewed by humans, structured annotations for page relationships, and specifications for navigation testing. It features a comprehensive five-axis evaluation that assesses build quality, navigation, visual accuracy, code maintainability, and efficiency. Additionally, the benchmark introduces state-isolated navigation testing to avoid cascading navigation failures and an anchor-reference method for visual fidelity. The findings are documented in a paper available on arXiv (arXiv:2607.28645).
Key facts
- MobileForge is the first benchmark for project-level multi-screen mobile app generation.
- It includes real mobile apps, human-reviewed screens, structured page-relationship annotations, and navigation test specifications.
- The benchmark evaluates five axes: build, navigation, visual fidelity, code maintainability, and efficiency.
- It introduces state-isolated navigation testing to avoid cascading failures.
- It proposes an anchor-reference method for visual fidelity evaluation.
- The paper is available on arXiv with identifier 2607.28645.
- Existing benchmarks are limited to single-page generation and cannot evaluate cross-page navigation.
- The research aims to improve the conversion of visual designs into executable code for mobile products.
Entities
Institutions
- arXiv