Anthropic says Claude Opus 5 is both its most capable and its cheapest to run

The variable that now governs enterprise AI deployment decisions is inference cost per output token, not benchmark rank. Benchmarks surface in the sales pitch, but token pricing is the numb