惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

V
Visual Studio Blog
罗磊的独立博客
小众软件
小众软件
T
Tailwind CSS Blog
宝玉的分享
宝玉的分享
博客园_首页
N
Netflix TechBlog - Medium
B
Blog
Recent Announcements
Recent Announcements
Y
Y Combinator Blog
Blog — PlanetScale
Blog — PlanetScale
L
LangChain Blog
F
Fortinet All Blogs
The GitHub Blog
The GitHub Blog
Stack Overflow Blog
Stack Overflow Blog
C
Check Point Blog
Last Week in AI
Last Week in AI
Jina AI
Jina AI
V
V2EX
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
博客园 - 叶小钗
博客园 - 【当耐特】

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
How to keep Claude Code on Opus 4.6 without breaking your...
Dimitar Stoy · 2026-05-14 · via DEV Community

Dimitar Stoyanov

There's been a lot of discussion recently around how Claude Code updates affect behavior, predictability, and token usage.

When Claude Code updates, it's not just the model that changes - the CLI harness, system prompts, tool-calling behavior, and context management all evolve too. Newer is often better. But not always for every workflow.

In my own setup, I noticed that some skills, agents, and project workflows I'd built behaved differently after recent updates. Not necessarily worse, but differently enough that predictability started to matter.

For AI-assisted development, that matters.

The problem

Sometimes you want the latest model. Sometimes you want the setup your workflow was built around.

There's no built-in way to run an older Claude Code version side-by-side with the latest one. If you pin the model with --model, you still get the latest harness — and harness changes can affect behavior just as much as model changes.

The solution: claude46

I made a small open-source helper that gives you a separate claude46 command:

  • Claude Code pinned to 2.1.110 (the pre-Opus-4.7 harness)
  • Opus 4.6 1M as the default model
  • Auto-updates fully blocked for that launcher only
  • Your normal claude install is left untouched and free to update

One-liner install

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/sparklingneuronics/claude-code-helpers/v0.2.0/install-claude46.sh | bash

Enter fullscreen mode Exit fullscreen mode

Windows (PowerShell):

irm https://raw.githubusercontent.com/sparklingneuronics/claude-code-helpers/v0.2.0/install-claude46.ps1 | iex

Enter fullscreen mode Exit fullscreen mode

Then just run claude46 instead of claude.

Is it better?

Not universally. I still use newer versions. But for workflows tuned around previous behavior — especially skills and agents built on top of Claude Code — having a stable pinned setup is useful.

The same idea as pinning a compiler or framework version: you want to choose when to upgrade, not have it happen under you.

How it works

The bash installer downloads Claude Code 2.1.110 to ~/.local/share/claude46/ and writes a wrapper script that sets DISABLE_AUTOUPDATER=1, DISABLE_UPDATES=1, and ANTHROPIC_MODEL=claude-opus-4-6[1m] before exec-ing the pinned binary.

The Windows installer does the same via npm install --prefix into %LOCALAPPDATA%\claude46\, writing a .cmd wrapper with the same env vars.

Both are tested end-to-end via CI on real runners.

Links