Claude Sonnet 5 Keeps Token Prices Despite Cost Doubling

The math is simple and brutal. You pay double for equivalent useful output. This sleight of hand works because Anthropic sells tokens, not intelligence.

If the new model burns more tokens to solve a problem—and THE DECODER confirms it does—your invoice inflates while that per-token rate sits perfectly, deceptively still. It's a shell game. The model's middling fifth-place rank almost fades next to this pricing architecture.

The fact that it sometimes beats the costlier Opus just heightens the strangeness. You're being charged more for a less expensive model to operate, because the metrics are gamed. Transparency in AI pricing was always fragile.

Now it's theater. The real cost hides in plain sight, shrouded by a cloud of identical, meaningless numbers.