Codex 5.3 vs Opus 4.6. The Better Model Doesn't Always Win.

Two SOTA models come out within minutes of each other, claim to be the best... I spent the weekend testing them both out. Codex 5.3 made a huge jump in performance and speed and is a better model than Opus 4.6. But alongside Opus 4.6, Claude Code's feature set continues to grow, making it the better overall product for getting real work done. In this video, I break down why the model matters less than the harness. I walk through Claude Code's newest features — Agent Teams, persistent memory management, the built-in task list, and the /insights command — and explain why these features keep me coming back to Claude Code over Codex CLI and the new Codex app. I also share my honest take on both models after spending time building with each, specifically native iOS/macOS apps where AI coding tools tend to struggle the most. ⏱️ TIMESTAMPS 0:00 – My honest take: Codex 5.3 vs Opus 4.6 1:00 – Why the harness matters more than the model 2:00 – Agent Teams: what subagents should have been 3:00 – How Agent Teams use Tasks and the mailbox 4:00 – The new Memory system (and why your CLAUDE.md is too long) 5:00 – How memory files are stored and the Git tradeoff 6:00 – /insights: Claude Code evaluates how YOU code 🔗 RESOURCES Claude Code Agent Teams Docs: https://code.claude.com/docs/en/agent-teams Claude Code Memory Docs: https://code.claude.com/docs/en/memory GPT-5.3 Codex System Card: https://openai.com/index/gpt-5-3-codex-system-card/ /insights Reddit post: https://x.com/trq212/status/2014480496013803643 GSD (Get Shit Done) Framework: https://github.com/glittercowboy/get-shit-done Book a call with me → https://yedatechs.com/#discovery-call Sponsorship inquiries → hi@yedatechs.com #ClaudeCode #Codex #Opus46 #AIcoding #AgentTeams