Simon Willison updated his Claude Token Counter tool to compare tokenization across Claude models, and the results reveal a quiet price hike. Opus 4.7 uses a new tokenizer that inflates token counts by 1.0 to 1.35x for text, per Anthropic's own numbers. Willison's testing found the real-world impact can be steeper. The Opus 4.7 system prompt consumed 1.46x more tokens than Opus 4.6. Since both models share identical per-token pricing ($5 per million input, $25 per million output), that token inflation translates to roughly 40% higher costs for typical text workloads. A text-heavy PDF showed a more modest 1.08x increase, so the hit depends on what you're sending.
Images are where it gets expensive. Opus 4.7 now supports images up to 2,576 pixels on the long edge, roughly 3.75 megapixels. That's more than triple the resolution of previous Claude models, and the token count reflects it. Willison tested a 3456x2234 pixel PNG and got a 3.01x token increase compared to Opus 4.6. But when he downsized to a 682x318 pixel image, the counts were nearly identical at 314 versus 310 tokens. You're paying for the model to actually see more detail.
Anthropic didn't exactly broadcast the tokenizer swap. The per-token pricing page shows the same numbers for Opus 4.7 as 4.6. The change that makes those tokens cost more? That's in the technical documentation. Developers migrating to Opus 4.7 should run their typical payloads through Willison's tool before committing. The tool now supports side-by-side comparisons across Opus 4.7, Opus 4.6, Sonnet 4.6, and Haiku 4.5.
Anthropic kept the per-token sticker price the same. Your actual bill won't reflect that.