Grid-Based Test Shows Gemini Outperforms GPT-5.4 on Dental Radiographs, but Both Lag Behind Students

In a grid-based assessment, Gemini 3.1 Pro outperformed GPT-5.4 on panoramic and periapical dental radiographs, while GPT-5.4 matched students on cephalometric landmarks. Both MLLMs lagged behind student consensus overall, with insufficient spatial reasoning for autonomous deplo…

• GPT-5.4 matched student-level performance on cephalometric landmarks but was outperformed by Gemini 3.1 Pro on panoramic and periapical tasks.
• Guided prompting caused heterogeneous effects: GPT-5.4 mislocalized the lower-left canine apex to the incorrect lower-right molar region, a prompt-resistant error absent in Gemini.
• The novel grid-based coordinate framework enables reproducible, modality-agnostic benchmarking of MLLM spatial reasoning.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.