Tag
The author argues that Pareto frontier benchmarks mislead general users, claiming that when pricing is compared via subscription rather than API rates (with cache hits), Opus 5.5 using Claude Code at 20x is superior to every other model.