Descriptions:
On the same day Anthropic released Claude Opus 4.6 and OpenAI followed roughly 18 minutes later with GPT-5.3 Codex, Greg Isenberg and engineer Morgan Linton — a former Sonos executive and active AI investor — put both models through a live head-to-head comparison centered on rebuilding Polymarket, the multi-billion dollar prediction market platform.
Linton opens with a practical setup guide for Opus 4.6 in Claude Code, covering the correct npm update sequence, how to navigate to the settings.json file in the .claude directory, and how to lock in the model version — mistakes many developers were making on launch day. The core of the video demonstrates Opus 4.6’s new agent teams feature, which automatically spawns parallel specialized sub-agents for architecture research, UX design, testing, and market analysis before writing a single line of code. GPT-5.3 Codex, by contrast, operates as a focused single agent that immediately begins scaffolding the codebase.
The comparison reveals a genuine methodological split: Opus 4.6 prioritizes parallel research and synthesis before implementation, while Codex prioritizes speed and production-code quality. Linton suggests Codex may have an edge for developers who want cleaner code fast, while Opus 4.6’s multi-agent coordination is better suited for complex projects that benefit from upfront architecture planning. A useful technical reference for developers choosing between the two models for serious engineering work.
📺 Source: Greg Isenberg · Published February 06, 2026
🏷️ Format: Comparison
