Benchmark RadarBIM-Edit
BIM-Edit evaluates large language models on natural-language editing of Industry Foundation Classes (IFC) building models. The benchmark includes 324 editing tasks across 11 realistic building models and 36 synthetic scenes. Tasks are categorized as direct, spatial, or topological instructions, and outputs are scored on geometric accuracy, semantic validity, and topological consistency.
- Released
- 2026-06-18
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Construction and architectural design rely on structured BIM models; the evaluation gap is that existing benchmarks mostly test geometry and creation from scratch. BIM-Edit measures scene understanding and semantic relational preservation, providing a capability signal for practical engineering workflows where editing is central.
Motivation
Large language models (LLMs) are increasingly applied to computer-aided design (CAD) to generate design artifacts from textual instructions.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.