Simon Willison @simonwillison.net · Apr 20 I upgraded my Claude token counter tool to compare different models and Opus 4.7 appears 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 simonwillison.net/2026/Apr/20/...
165 likes 11 replies
Replies
LOL
3x on images is wild. same price per token but triple the consumption is effectively a 3x price hike on vision tasks that they just... didn't mention?
Token based pricing is an absolute disaster for consumers. Its full of hidden price charges / rises and the potential for over billing which are then near impossible to prove afterwards.
Important to note: that 3x increase for images is entirely due to Opus 4.7 being able to handle higher resolutions. I tried that again with a 682x318 pixel image and it took 314 tokens with Opus 4.7 and 310 with Opus 4.6, so effectively the same cost.
What does it mean that 4.7 uses more tokens for text? How do you count them?
This is a fascinating measurement. The compounding effect of this on agentic workloads (tokenizer + thinking + verbosity) turns a flat price into a 1.5–2x effective cost delta. The denominator shift is much quieter than a price hike.
So it looks like token-saving approaches like jCodeMunch and caveman are now even more relevant than before.
Per-token parity is the trick. More tokens at the same rate is a quiet ~46% price bump for text workloads without the sticker shock. 3x on images is the wild one. Worth flagging in vision-heavy pipelines before bills spike.
If you downsample the image identically to what opus 4.6 would have done, is the image token usage equal between models? Feels way too linear that you see a 3x token increase when they increased the allowable resolution 3x.
why use many token when few do trick