A recent evaluation pitted four local coding models against each other in generating and modifying Three.js scenes. Two independent Claude Sonnet graders assessed the outputs, scoring them on a scale of 0 to 3. The gpt-oss-20b model emerged as the top performer, achieving a total score of 43.5 out of 72, while Qwen3-Coder-30B-A3B followed with 28.0. Notably, many models struggled with functional code, with some breaking during execution or hitting token limits without producing any code. AI
IMPACT Provides insights into the comparative performance of local coding models for 3D scene generation, aiding developers in model selection.
RANK_REASON The cluster describes an independent evaluation of multiple AI models on a specific task, which falls under research. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
