Simon Willison upgraded his Claude Token Counter tool this week to compare token counts across Anthropic's models, and the results reveal a quiet pricing shift. Claude Opus 4.7 uses a new tokenizer that maps the same text to more tokens. Anthropic said the increase would be 1.0 to 1.35x depending on content. But when Willison tested the Opus 4.7 system prompt, he found it consumed 1.46x the tokens compared to Opus 4.6. Same per-token price, more tokens per request. That works out to roughly 40% more expensive for identical workloads. The sticker shock gets worse with images. Opus 4.7 supports higher resolution inputs, up to 2,576 pixels on the long edge. A 3456x2234 pixel PNG cost 3.01x more tokens on 4.7 than on 4.6. The increase comes from the higher resolution support, not the tokenizer itself. When Willison tested a smaller 682x318 pixel image, the difference was negligible (314 vs 310 tokens). A 15MB text-heavy PDF showed only a 1.08x multiplier. So the real cost impact depends heavily on what you're feeding the model. Anyone building agents or pipelines with Opus 4.7 should pay attention. If your workload is mostly text prompts and system instructions, budget for that 40% bump. If you're processing high-resolution images, the increase could be much steeper. Willison's tool is worth bookmarking if you want to audit your actual costs before committing to the upgrade.