InsightBench

Findings

• IC-312GPT-4o achieves the highest insight-level Llama-3-eval score (0.60) among all tested LLM backbones on InsightBench multi-step data analytics [eval]
• IC-313All four LLM backbones fail to detect a planted linear trend in incident resolution time when the slope is below 0.1, and detection rates diverge sharply above that threshold [eval]
• IC-314Llama-3-70b as an LLM-based evaluator (Llama-3-eval) produces agent rankings consistent with GPT-4-based G-Eval on InsightBench [eval]