Claude Opus 4.7

Re:Claude Opus 4.7 by XCSme on Thursday April 16, 2026 @08:37PM (#47801254)

The reasoning modes are really weird with 4.7

In my tests, asking for "none" reasoning resulted in higher costs than asking for "medium" reasoning...

Also, "medium" reasoning only had 1/10 of the reasoning tokens 4.6 used to have.

[ Parent | on HN ]

Re:Claude Opus 4.7 by nisarg2 on Thursday April 16, 2026 @11:21PM (#47802137)

Insane! Even Haiku doesn't make such mistakes.

[ Parent | on HN ]

Re:Claude Opus 4.7 by XCSme on Friday April 17, 2026 @06:36AM (#47804437)

I am not sure it's a mistake, this might be their new "adaptive reasoning" + hidden reasoning trace, so we can't verify.

[ Parent | on HN ]

Re:Claude Opus 4.7 by XCSme on Thursday April 16, 2026 @08:38PM (#47801261)

Oh, and also, the "none" and "medium" variants performed the same (??)

[ Parent | on HN ]

Re:Claude Opus 4.7 by amelius on Friday April 17, 2026 @03:48AM (#47803486)

Claude is known for its shitty metering.

[ Parent | on HN ]

Re:Claude Opus 4.7 by Eifert on Saturday April 18, 2026 @07:10AM (#47814934)

Medium reasoning has regressed since 4.6. While None and Max have improved since 4.6 in our benchmark. We suspect that this is how Claude tries to cope with the increased user base. Note, Google and OpenAI probably did something similar long ago.

[ Parent | on HN ]