ScienceVision tests models’ 3D scientific reasoning
ScienceVision evaluates AI models across biology, chemistry, medical imaging, architecture, and mechanical CAD using 250 questions built from real 3D data. According to 新智元, tasks require models to inspect, measure, and reason about 3D objects, with deterministic scoring and expert-authored reference answers. The reported top score was 0.53 from GPT-6 Astra with Codex, while no model led across all five fields.
The benchmark provides a unified test of whether models can apply domain knowledge to genuine 3D evidence rather than merely produce visually convincing CAD.
ASI启示录 2026-10-30 19:56 北京
AI会画3D，却未必看得懂3D科学世界。
新智元报道
前一阵，GPT-6 Astra 画 CAD 的视频在 X 上实火。
比如这个涡轮增压器，每个零件都单独建模，不仅能正常转动，还能一键原地拆解展开，视觉效果相当炸裂。
#ScienceVision#UniPat AI#Elorian AI#GPT-6 Astra#Codex#Claude Fable 5.1#Claude Opus 5.5#Gemini 4 Argon#Qwen3.8 Max#SciVisAgentBench