Autor: MedaBeda ++++
Datum: 16. 04. 2026 20:53:21
https://www.reddit.com/r/ClaudeAI/comments/1sn57af/introducing_claude_opus_47_our_most_capable_opus/
ClaudeOfficial
Introducing Claude Opus 4.7, our most capable Opus model yet.
------------
ClaudeAI-mod-bot
MOD •
4h ago
•
Stickied comment
• Edited 1h ago
TL;DR of the discussion generated automatically after 400 comments.
The community is overwhelmingly skeptical, believing Opus 4.7 is just the pre-nerfed version of 4.6 being re-released at a higher effective cost. You'll likely burn through your weekly usage limit just reading this post, as that's the number one complaint in this thread.
Here's the breakdown of the drama:
Usage Limits & Cost: The new tokenizer uses up to 35% more tokens for the same input, which everyone sees as a stealth price hike. Users on all plans are reporting hitting their 5-hour and weekly limits after just a handful of prompts.
Major Regression: A highly-upvoted comment points out that long-context retrieval (MRCR benchmark) has tanked from 78.3% in 4.6 to just 32.2% in 4.7. An Anthropic dev explained they're phasing out this benchmark, but users who rely on long context are seriously concerned.
Claude Code is Busted: Multiple users report that 4.7 is on a hair trigger, constantly flagging benign, simple code as "malware" and refusing to perform edits.
Downgraded Features: The "Extended Thinking" toggle in the web app has been removed and replaced with "Adaptive Thinking," which you can't control.
The Mythos Tease: People are getting really tired of Anthropic including the unreleased "god model" Mythos in benchmark charts, seeing it as a way to distract from 4.7's regressions and inferiority to competitors like GPT-5.4 in some areas.
@prog: |
|
Autor: Anonymni ++++
