Member-only story
Sonnet 5 launched with a promotional rate of $2/$10 per MTok that expires August 31, 2026 #
The three-part trap, assembled #
Component one: the promotional rate expires.
Sonnet 5 launches at $2 per million input tokens and $10 per million output tokens through August 31, 2026. On September 1, those rates become $3 input and $15 output. That is a clean, hard date with no auto-extension. The jump is a clean +50% on every token.
Component two: the tokenizer inflates your counts.
Anthropic’s announcement explained that Sonnet 5 uses an updated tokenizer that changes how the model processes text to improve performance, and that the same input can map to roughly 1.0–1.35x more tokens depending on content type. This is not a vague estimate applied to exotic workloads. The effect is heaviest on code, structured data, and non-English text. For plain English prose it may be close to 1.05x. For a code-heavy or multilingual workload it can reach 1.35x. The only way to know where your workload lands is to measure it directly using the token-counting API.
One downstream effect worth noting: although the 1M-token context window is unchanged, the absolute volume of readable text…