TIDE: Tree-Induced Fine-Grained Diagnostic Evaluation from Model Behaviors
Claude Opus 4.6 led GPT-5.4 on microeconomic market reasoning but trailed it on contract law—contrasts hidden by aggregate scores.
The authors tested TIDE on five models across four benchmarks. It builds a hierarchical capability tree from benchmark samples and model scores, using an LLM judge to propose and classify groups and behavior-based information gain to select splits. Claude scored 100% versus GPT-5.4’s 83.3% on microeconomic market reasoning; on contract law reasoning, the scores were 77.8% and 88.9%, respectively.
ArXiv · AI/CL/LG's note
The authors tested TIDE on five models across four benchmarks. It builds a hierarchical capability tree from benchmark samples and model scores, using an LLM judge to propose and classify groups and behavior-based information gain to select splits. Claude scored 100% versus GPT-5.4’s 83.3% on microeconomic market reasoning; on contract law reasoning, the scores were 77.8% and 88.9%, respectively.
ArXiv · AI/CL/LG's note
score 4