THE SWITCH

Switch Claude Code back to Opus 4.6.

I ran Opus 5 and Opus 4.6 side by side for two weeks. 4.6 follows instructions, stays on task, and a three-line function stays a three-line function. The switch is one line, and it's on this page.

Neej

Neej

I build AI teams that run real businesses

Work with meBuilt for your business, or coached to build it yourself

WHAT'S GOING ON

  • →Opus 5 tops the charts. BenchLM scores it 59 on the Intelligence Index against 44 for 4.6. VentureBeat's read after a month of real usage: raw benchmark scores don't predict the bill.
  • →The complaints pile up in the same two buckets: it over-engineers simple tasks and it ignores your instructions. One 82K-view breakdown called Opus 5 'the beginning of Anthropic's downfall.' Devs on X are quietly moving back to older versions.
  • →My two weeks: 4.6 just does what I tell it. Fewer tokens, fewer corrections, smaller diffs.

Drop your email and I'll send my full Claude Code setup

The model-switch prompt plus the config I actually run: the settings that keep agents on task and the checks I do before trusting a model with real work. Plus what I learn shipping with Claude Code every week.

Join 3,000+ builders already getting this. No spam.

Two ways in: I build it for your business, or I teach you to build it yourself. Owners, operators and learners all welcome. Start here.

Work with me

My tool stack

Every AI tool I actually run, with my referral links and the setup notes that go with them.

referral codes · setup notes

Neej

Follow the build

I post the whole build in public, broken parts included. A new system like this every week.

@buildwithneej on instagram

also on