Simon Willison Finds Claude Opus 4.7 Tokenizer Raises Real API Costs

Simon WillisonSimon Willison

· Updated

Simon Willison analyzed the new Claude Opus 4.7 tokenizer using his custom counter tool and found significant token inflation. While Anthropic kept per-token pricing the same, the increased token count for text and images creates a hidden price hike for API users.

Simon Willison, creator of Datasette, updated his token counter tool to compare the new Opus 4.7 against Opus 4.6. His testing revealed that the model uses 1.46x more tokens for raw text and up to 3x more for images. A 30-page PDF test showed a smaller 1.08x increase.
Text token multiplier
1.46x
Image token multiplier
Up to 3x
PDF token multiplier
1.08x
Models compared
Opus 4.7 vs Opus 4.6
Analysis tool
Claude Token Counter

This finding highlights a hidden price hike for API users. Although Anthropic maintained static per-token rates, the increased token volume raises total costs. This is the opposite of the tiered cost-saving approach seen at Google, illustrating a divergence in how labs manage high-reasoning model economics.

Developers should recalibrate API budgets for this inflation when migrating to the recently released Opus 4.7. While Anthropic increased subscriber rate limits to offset costs for chat users, API developers must use Simon's tool to estimate the impact, as PDF results showed much lower inflation.

View the full update on tools.simonwillison.net
Simon Willison
Simon Willison
@simonw
X

I upgraded my Claude token counter tool to compare different models and Opus 4.7 does appear to use 1.46x times the tokens for text and up to 3x the tokens for images - it's priced the same as Opus 4.6 on a per-token basis so this is actually a pretty big price bump https://t.co/PPGrwHf39D

128retweets1.5klikes
View on X

Still wondering? A few quick answers below.

Simon Willison discovered that Claude Opus 4.7 uses significantly more tokens than Opus 4.6 for the same input. His testing showed a 1.46x increase for text and up to 3x for images. This means that even though the price per token is the same, the actual cost to process data has increased.

The cost increase is caused by token inflation, where the model's tokenizer breaks text and images into more numerical units than before. Since Anthropic bills per token, using 1.46 times more tokens for the same text results in a 46 percent price hike for API users despite the static per-token rate.

Developers should use Simon Willison's token counter tool to estimate the specific impact on their data types. While text and images see high inflation, text-heavy PDFs showed a much lower 1.08x increase. Recalibrating API budgets is essential before migrating production workflows to the new model to avoid unexpected billing spikes.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update