Compare

Claude Opus 5 vs Claude Sonnet 4.6

Input, output, and cached reads all scale by the same multiple between these two, so Claude Sonnet 4.6 comes out 1.7× cheaper on every shape below — not just some of them. No traffic pattern flips this one.

Per million tokens

Claude Opus 5

Claude Sonnet 4.6

Input$5.00$3.00
Output$25.00$15.00
Cached input$0.50$0.30
Context1M1M
Max output128K128K
ProviderAnthropicAnthropic

Monthly cost by workload

Chat assistant

60,000 requests · 1,200 in / 400 out · 30% cached — short prompts, short answers, high volume

Claude Opus 5

$862.80

Claude Sonnet 4.6

$517.68

Claude Sonnet 4.6 is 1.7× cheaper on this shape — $345.12 a month.

RAG over documents

20,000 requests · 18,000 in / 700 out · 75% cached — large retrieved context, heavily cached prefix

Claude Opus 5

$935.00

Claude Sonnet 4.6

$561.00

Claude Sonnet 4.6 is 1.7× cheaper on this shape — $374.00 a month.

Coding agent

4,000 requests · 45,000 in / 6,000 out · 80% cached — long tool-calling loops, big cached prefix

Claude Opus 5

$852.00

Claude Sonnet 4.6

$511.20

Claude Sonnet 4.6 is 1.7× cheaper on this shape — $340.80 a month.

Run these numbers on your own workload