IndustryForge-27B is built on Qwen3.5-VL-27B and trained on approximately 52,000 multimodal industrial-CAD samples covering CAD visual QA, parametric CAD code, assembly code, and Windows COM automation for Inventor and SolidWorks. The paper reports an average gain of 33.65 percentage points over the base model across four CAD benchmarks and claims that it outperforms GPT-5.4 on all four. Across eleven general-capability benchmarks, it reportedly improves the base model by 1.56 points on average without catastrophic forgetting.
No heat snapshots are available in the last 24 hours.