Eating Up 35% More Tokens
Key point
Because of the new tokenizer in Opus 4.7, the same input burns through tokens up to 35% faster than 4.6.
Details
With the new tokenizer in Opus 4.7, the same text now uses about 1x~1.35x, up to 35% more tokens, compared to 4.6.
So if you keep using context files and work habits built around 4.6 as they are, your session gets used up much faster than it feels like it should. The recent reports of "a single prompt killed the session" are explained as being tied to this change.
One user who experienced this firsthand said that even after upgrading to Max 5x, just doing ordinary work hit 100% session usage within a single work block. The work involved was at the level of tidying up a workspace, writing an SOP, a small internal web app, and organizing markdown context files.
Responses that actually helped include the following.
- Trimming project files down to about one page
- Starting new work in a new chat
- Not pasting the same document twice
- Organizing the prompt in a notepad first before sending it
- Replacing "rewrite the whole document" requests with diffs or specific edit requests
The key point is that on Opus 4.7, the token cost is higher for the same amount of work, so context management and request granularity need to be handled more conservatively.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.