Token Comparison

Claude Token Counter has been updated to let you put the same input into multiple Claude models and compare token counts.

Claude Opus 4.7 appears to be the first model with a changed tokenizer, so the comparison is effectively most meaningful between 4.7 and 4.6. However, since Claude's token counting API can accept any Claude model ID, the tool currently includes all the major models: Opus 4.7, Opus 4.6, Sonnet 4.6, and Haiku 4.5.

Anthropic explained that Opus 4.7's updated tokenizer can turn the same input into roughly 1.0 to 1.35x more tokens. When Simon Willison tested it with the Opus 4.7 system prompt, 4.7 used 1.46x more tokens than 4.6, and even though the pricing is the same, actual costs could end up about 40% more expensive.

The difference was even larger for image input. When feeding in a 3456×2234 pixel, 3.7MB PNG, Opus 4.7 used 3.01x the tokens compared to Opus 4.6.

Opus 4.7 also comes with enhanced high-resolution image processing. It can accept images up to 2,576px on the long side, which is more than 3x higher than previous Claude models.