GPT-5 Codex vs Claude Sonnet 4.5 vs Grok Code vs Composer (Speed Test)

Description

🤖 I put four AI coding models head-to-head with identical prompts to build a habit tracking app. Composer, Claude Sonnet 4.5, Grok Code, and GPT-5 Codex. Same task, same requirements, wildly different results and build times. One model finished in 90 seconds. Another took 26 minutes. But speed isn't everything - the quality, features, and functionality varied dramatically across all four builds. ⏰ TIMESTAMPS: 00:00 Introduction 00:47 The prompt that we will be giving each model 03:54 Watching all four agents build in parallel 04:41 Reviewing the time each model took to complete the build 05:38 Reviewing the total code and architecture 06:21 Testing Composer's Habit Tracker App 09:19 Testing Claude's Habit Tracker App 13:06 Testing Grok's Habit Tracker App 17:21 Testing Codex's Habit Tracker App 22:59 Final thoughts on each model's performance 24:53 Share your thoughts and request test ideas in the comments 🎯 THE TEST: Task: Build a habit tracking app with React Requirements: Add habits, track daily completion, display streaks, persist data, creative freedom for additional features Conditions: Identical prompts, isolated directories, one-shot build (no iterations) Models Tested: Composer 1 (Cursor) Claude Sonnet 4.5 Grok Code GPT-5 Codex 🚀 WHAT I TESTED: Build Performance: Time to completion Lines of code efficiency File architecture decisions Feature creativity and implementation App Quality: Core functionality (habit tracking, streaks, data persistence) UI/UX design and polish Additional features implemented Bug-free execution Mobile responsiveness Real-World Usability: Calendar views and date tracking Progress tracking accuracy Dark mode implementation Data management (edit, delete, archive) 📊 KEY RESULTS: Build Times: Composer: 1 minute 35 seconds (fastest) Claude: 5 minutes 5 seconds Grok: 11 minutes 43 seconds Codex: 26 minutes 23 seconds (slowest - 17x longer than Composer!) Code Efficiency: Composer: 10 files, 1,093 lines Claude: 16 files, 1,578 lines Grok: 15 files, 1,570 lines Codex: 19 files, 2,082 lines (most code) Feature Comparison: Composer: Basic functionality, minimal features, progress tracking bugs Claude: Calendar view, motivational quotes, categories, 14-day view, some tracking bugs Grok: Similar to Claude, calendar view, category filters, formatting issues Codex: Most features - monthly calendar, focus mode, reminders, "why this matters" field, archive system, detailed analytics Critical Issues Found: Composer: 700% progress bug, weekly-based instead of daily tracking Claude: Today's progress requires 14-day completion to update Grok: Date formatting errors, completed today not updating correctly Codex: Edit button UX issue, unclear current day indicator 💡 TRADE-OFFS: Speed vs Features: Composer delivered fastest but most basic app Codex took 17x longer but built most comprehensive solution Claude and Grok balanced speed with quality (middle ground) Code Efficiency: Composer used half the code of Codex More code doesn't always mean better results File organization varied significantly Use Case Considerations: Quick iteration workflow: Composer excels Feature-rich one-shot builds: Codex delivers Balanced approach: Claude or Grok ⚠️ METHODOLOGY NOTES: Used Cursor 2.0 agents running in parallel Identical detailed prompts with creative freedom No debugging or follow-up prompts (pure one-shot test) All models given same requirements and constraints Tested actual functionality, not just visual design 🔗 RESOURCES: Cursor IDE: https://cursor.sh Sign up for updates: https://snapperai.io 📺 RELATED VIDEOS: GPT-5 Codex vs Claude Sonnet 4.5 (Clear Winner): https://www.youtube.com/watch?v=xMHQDiSAPuo GPT-5 Codex in Cursor: Complete Setup & Tutorial Guide: https://www.youtube.com/watch?v=1wP97Rd62cg 🎯 PERFECT FOR: Developers choosing between AI coding assistants Teams evaluating model performance for specific tasks Cursor 2.0 users exploring multi-agent workflows Anyone comparing Composer, Claude, Grok, and Codex Developers interested in build speed vs quality tradeoffs Vibe coders looking for the right model for their workflow 💬 YOUR EXPERIENCE? Which of these models do you prefer for building apps? Have you tested Cursor 2.0's parallel agents? What workflows or tasks should I test next? Drop your experience and requests in the comments! 🎁 STAY CONNECTED: 👉 SUBSCRIBE for more AI coding comparisons and real-world tests 👉 Newsletter & Resources: https://snapperai.io