Simon Willison • 4/20/2026

Claude Token Counter, now with model comparisons

This article details an upgrade to the Claude Token Counter tool, now supporting token count comparisons across different Claude models. It highlights that Opus 4.7 uses an updated tokenizer, resulting in 1.0–1.35× more tokens for text and up to 3× more for high-resolution images, leading to roughly 40% higher costs. The tool accepts images and PDFs, with tests showing varying token multipliers. The article is a technical update relevant to AI/LLM tokenization and cost analysis.

0 comments

#api #Anthropic #Model Comparison