AI coding token use varied 70-fold with the same model, The New Stack reports
The New Stack says agent harnesses—the software around a model—can change coding-agent costs by multiples. Its report uses three benchmarks to examine where extra tokens come from.
TLDR
The New Stack reports that Aider, Claude Code and OpenClaw ran an identical model, yet token use varied 70-fold. The outlet examines how different agent harnesses can produce dramatically different token usage, making model choice only part of the coding-cost equation.
Combined views
528
1 Source, first seen ago
1 likes1 saves1 reposts