SolidWorks-Bench: no AI model can assemble a single machine; Fable 5.2 leads at 57.3%

DevvMandal · x · 2026-11-19

SolidWorks-Bench tests whether AI models can handle real mechanical engineering tasks in SolidWorks. Models get close on individual parts but fail at assembly—no model completes a single machine. Fable 5.2 leads at 57.3% mean reward, ahead of GPT-6 Astra (42.6%), Sonnet 5.5 (34.7%), Opus 5.5 (34.2%), Kimi K3.1 (19.6%), DeepSeek V4.1 (12.7%), GLM 5.3 (7.3%), and MiniMax M3.1 (2.4%).

Related event: SolidWorks-Bench: No AI Can Assemble a Full Machine(2 posts)→

Original post →

More from Models

Models channel →