ResearchBenchmarksCorrelation with Human Likert Scores on FinePhyEvalFollow0.478Pearson's rPQSG0.414560.431030.44750.46397Jun 24, 2026Evaluation ResultsMethodMethodLinksPearson's rKendall's τSpearman's ρPQSGLLM Backbone=GPT-5.5LLM Backbone=GPT-5.52026.060.4780.3360.456PQSGLLM Backbone=Gemini-2....LLM Backbone=Gemini-2.5-Pro2026.060.4670.3060.406PQSGLLM Backbone=GPT-5LLM Backbone=GPT-52026.060.430.2630.366PQSGLLM Backbone=GPT-5.4LLM Backbone=GPT-5.42026.060.4290.2740.369PQSGLLM Backbone=Gemini-2....LLM Backbone=Gemini-2.5-Flash2026.060.4170.2590.354