Coding agent scores: why 83.4% hides real costs
The Artificial Analysis Coding Agent Index v1.1 reveals how an 83.4% TerminalBench score masks variance in token efficiency and tool use.
Priya Nair
13 min read