terminal-bench

Augment Code benchmarked its Auggie agent against Claude Code on Claude Opus 4.7, reporting a 67.4% pass rate versus 66.3% while cutting costs by 33%. The company attributes savings to a semantic context engine that reduces cache read tokens by 32% and output tokens by 37% compared to Claude Code's keyword-based retrieval.