DEV Community: Jitendra Kumar

Stop Using Opus for Everything: The Claude Code /model Framework

Jitendra Kumar — Fri, 08 May 2026 17:11:02 +0000

https://youtu.be/S-2gZNJ4nGo

description: "A simple decision tree for switching between Haiku, Sonnet, and Opus in Claude Code. Includes the 3rd hidden model most developers don't know about. Day 4 of 10."
tags: claudecode, ai, productivity, vscode
cover_image:

series: 10 Days of Claude Code

TL;DR — Claude Code has THREE models, not two. Default to Sonnet. Use Haiku for mechanical tasks (renaming, comments, formatting). Switch to Opus only when Sonnet gets stuck. This framework alone tripled my Pro plan's lifespan.

Welcome to Day 4 of my "10 Days of Claude Code" series! 🚀

Catch up on previous days:

📌 Day 1 — /init to onboard Claude to your codebase
📌 Day 2 — /review and /ask for instant code audits
📌 Day 3 — /clear vs /compact for context management

Today: the command that most developers misuse — /model.

The Mistake Almost Every Pro User Makes

You opened Claude Code. You're on the Pro plan. You want the best results, so you set the model to Opus and forget about it.

Two hours later: usage limit hit.

Sound familiar? You're not alone. Defaulting to Opus is the #1 reason Pro plans run out fast.

The fix: actually use the /model command.

The Three Models — And Their Real Jobs

/model

This shows you the model picker. There are three options:

🟡 Haiku — The Speed Layer

/model haiku

The model nobody talks about. Lightning fast. Near-zero token cost.

Use it for:

✅ Renaming variables
✅ Writing JSDoc / docstrings
✅ Adding console.log statements
✅ Quick formatting fixes
✅ Simple search-and-replace
✅ Import additions

If your task is mechanical (the answer doesn't require reasoning), Haiku is the right tool.

🟠 Sonnet — The Workhorse

/model sonnet

This should be your default. Period.

Use it for:

✅ Bug fixes
✅ Refactoring
✅ Writing new functions from clear specs
✅ Multi-step file edits
✅ Code reviews
✅ Test writing
✅ Daily coding tasks

Sonnet handles 80% of what you actually do at a fraction of Opus's token cost.

🟣 Opus — The Heavy Hitter

/model opus

The most capable model — and the most expensive in terms of usage. Reserve it for when it earns its place.

Use it for:

✅ Architectural decisions
✅ Multi-file refactors with cascading impact
✅ Performance optimization
✅ Complex debugging where Sonnet got stuck
✅ Security analysis
✅ System design

If a task rewards getting it right the first time, that's Opus territory.

The Decision Tree I Actually Use

1. Start session → Sonnet (always)

2. Mechanical task?
   YES → Switch to Haiku
   NO  → Stay on Sonnet

3. Sonnet struggling 2x in a row?
   YES → Switch to Opus
   NO  → Stay on Sonnet

4. Solved the hard problem?
   YES → Switch back to Sonnet immediately

That's it. Four rules. No spreadsheet.

The Comparison Table

Task	Best Model	Why
Rename a variable	Haiku	Mechanical
Write a JSDoc block	Haiku	Pattern-based
Fix a TypeError	Sonnet	Needs context
Add a new feature	Sonnet	Multi-step logic
Refactor a class	Sonnet	Daily refactor work
Design a microservice	Opus	Architectural depth
Debug a race condition	Opus	Complex reasoning
Optimize an algorithm	Opus	Tradeoff analysis

Real-World Numbers

Before this framework:

⏱️ Pro plan lasted ~2-3 hours of heavy use
💸 Most days I hit usage limit before lunch

After this framework:

⏱️ Pro plan lasts the full workday
💸 On normal days, I have headroom to spare

Same code shipped. Same productivity. 3x more runway.

Try It in 30 Seconds

# In Claude Code
/model

# Pick one based on your current task
/model haiku   # for mechanical work
/model sonnet  # for daily coding
/model opus    # for hard problems

You can switch mid-session. You can switch as often as you want. The model just changes for the next prompt.

🎬 Watch the 60-Second Version

If you prefer video, I made a quick reel covering this:

📺 YouTube Short:

What's Next — Day 5

Tomorrow I'm covering the command that lets you track your Pro plan usage in real-time. Most developers don't know this dashboard even exists. It's the natural follow-up to Day 4 — once you switch models, you need to verify it's actually saving you tokens.

🔔 Follow me on Dev.to so you don't miss it!

💬 Your Turn

Did you know /model haiku existed?
Which model do you default to today?
Will you try the framework above?

Drop a comment — I read every single one. 👇

About the series: I'm sharing one Claude Code productivity tip every day for 10 days. Follow along on:

📺 YouTube
📷 Instagram
📝 Medium
💼 LinkedIn

Let's master Claude Code together! 🚀

ClaudeCode #AI #Productivity #VSCode #DevTools

Claude Code Slow? Master /clear and /compact in 2 Minutes (Day 3 of 10)

Jitendra Kumar — Wed, 06 May 2026 17:12:41 +0000

TL;DR — Long Claude Code conversations slow down because the context window fills up. Use /clear when starting a brand new task, and /compact when you want to keep working on the same task but free up memory. Master this rule and your Pro plan will go 2x further.

Welcome back to Day 3 of my "10 Days of Claude Code" series! 🚀

If you missed the earlier days:

📌 Day 1 — How /init instantly understands your entire project
📌 Day 2 — /review and /ask for instant code audits

Today we're solving a problem every Claude Code user faces but most don't know how to fix.

😩 The Problem Nobody Talks About

You start a Claude Code session in VS Code. Everything is fast and crisp. You build a feature, fix a bug, refactor some code.

Then 30 minutes in...

⏳ Replies become slower
💸 Tokens deplete faster
🤔 Claude starts giving slightly off answers
😤 You're getting frustrated

What's happening?

Every message you send adds to Claude's context window. Your old chat from 30 minutes ago — the one about a totally different file — is still being processed with every new request. It's like asking your friend a quick question while they're holding 200 sticky notes from earlier in the day.

Two slash commands solve this completely.

⚡ Command #1: `/clear`

What it does: Wipes the entire conversation history and starts fresh. Same session, same project, but Claude forgets everything you've discussed.

When to use it:

✅ Starting a completely new task (e.g., done with auth, now working on payments)
✅ Switching to an unrelated file or feature
✅ Claude is giving weird/confused responses
✅ You want maximum speed and a clean slate

How to use it:

/clear

That's it. One command. Fresh terminal, fresh brain, full speed.

Real example from my workflow:

[Worked on user authentication for 20 min]
[Bug fixed, feature shipped ✅]

Me: /clear

[New conversation begins]
Me: Now help me build the Stripe payment integration.

Claude doesn't carry over the auth context — which is exactly what I want. Less noise, faster response.

🗜️ Command #2: `/compact`

What it does: Summarizes your current conversation, keeps the important parts, and discards the fluff. You don't lose context — you compress it.

When to use it:

✅ Continuing the same task but the chat is getting long
✅ You're deep in a complex feature and need Claude to remember decisions
✅ You see token usage climbing but can't afford to lose context
✅ Mid-debugging session where history matters

How to use it:

/compact

Claude generates a summary of what you've discussed and replaces the long history with that summary. Same context, way less memory.

Real example:

[60 messages deep into building a dashboard]
[Multiple component decisions, design choices made]

Me: /compact

Claude: [summarizes the entire session into ~5 key points]

Me: Now add a filter dropdown to the chart we built earlier.

Claude still remembers the chart, the design system, the framework choice — but it's not re-reading 60 messages every time.

🎯 The Golden Rule (memorize this)

Situation	Command
Starting a new task	`/clear`
Continuing the same task	`/compact`
Claude is confused	`/clear`
Conversation is long but relevant	`/compact`

That's it. Two commands, infinite productivity.

💰 Why This Matters for Pro Users

If you're on the Pro plan (like me), this is money saved.

Every token in your context window counts toward your usage
Bloated conversations = wasted Pro quota
A 5-second /clear or /compact can save you hundreds of tokens per task

I personally run /clear between every major task switch. My Pro plan now lasts way longer than it did before I knew this.

🧪 Try It Right Now

Open Claude Code in VS Code, run a few prompts, then try:

Type /compact — watch the conversation collapse into a clean summary
Type /clear — see a fresh terminal, ready for a new task

You'll feel the difference immediately.

🎬 Watch the 60-Second Version

If you prefer video, I made a quick reel covering this:

📺 YouTube Shorts: https://www.youtube.com/shorts/lM1YE2FuD8o
📱 **Instagram Reel: **https://www.instagram.com/tech_jitendra_nadol/reel/DX9oSwNMhIh/

📅 What's Next?

Tomorrow — Day 4 I'm covering the /model command — when to use Sonnet vs Opus and how to save tokens by picking the right model for the right job. This one's a game-changer for Pro users.

🔔 Follow me on Dev.to so you don't miss it!

💬 Your Turn

Did you know about /clear and /compact?
Which one will you use more often?
What's your biggest Claude Code pain point?

Drop a comment — I read every single one. 👇

About the series: I'm sharing one Claude Code productivity tip every day for 10 days. Follow along on:

🐦 Twitter/X: https://x.com/JitendraV9521
📷 Instagram: https://www.instagram.com/tech_jitendra_nadol/
📺 YouTube: https://www.youtube.com/@techjitendranadol9876

Let's master Claude Code together! 🚀

ClaudeCode #AI #Productivity #VSCode #DevTools

Yajra DataTables Part 2 is LIVE!

Jitendra Kumar — Thu, 22 Jan 2026 17:38:10 +0000

Video
Today I explained the NEW DataTable Class approach
(query & dataTable methods) in Laravel 12.

Next video: Action Buttons (Edit/Delete) 🔥

Laravel 12 Yajra DataTables – New Way Explained

Jitendra Kumar — Mon, 19 Jan 2026 19:28:43 +0000

Video

Laravel 12 Yajra DataTables – New Way Explained

Jitendra Kumar — Mon, 19 Jan 2026 19:28:43 +0000

Video
https://youtu.be/yhhfc7Xy4-Q

Laravel 12 Yajra DataTables – New Way Explained

Jitendra Kumar — Mon, 19 Jan 2026 19:28:43 +0000

Video

DEV Community: Jitendra Kumar

Stop Using Opus for Everything: The Claude Code /model Framework

https://youtu.be/S-2gZNJ4nGo

series: 10 Days of Claude Code

The Mistake Almost Every Pro User Makes

The Three Models — And Their Real Jobs

🟡 Haiku — The Speed Layer

🟠 Sonnet — The Workhorse

🟣 Opus — The Heavy Hitter

The Decision Tree I Actually Use

The Comparison Table

Real-World Numbers

Try It in 30 Seconds

🎬 Watch the 60-Second Version

What's Next — Day 5

💬 Your Turn

ClaudeCode #AI #Productivity #VSCode #DevTools

Claude Code Slow? Master /clear and /compact in 2 Minutes (Day 3 of 10)

😩 The Problem Nobody Talks About

⚡ Command #1: /clear

🗜️ Command #2: /compact

🎯 The Golden Rule (memorize this)

💰 Why This Matters for Pro Users

🧪 Try It Right Now

🎬 Watch the 60-Second Version

📅 What's Next?

💬 Your Turn

ClaudeCode #AI #Productivity #VSCode #DevTools

Yajra DataTables Part 2 is LIVE!

Laravel 12 Yajra DataTables – New Way Explained

Laravel 12 Yajra DataTables – New Way Explained

Laravel 12 Yajra DataTables – New Way Explained

⚡ Command #1: `/clear`

🗜️ Command #2: `/compact`