Summary
OpenAI recently merged its Codex platform with ChatGPT while simultaneously releasing three new models — GPT-5.6 Soul, Terra, and Luna — marking a significant shift in how the company positions its AI offerings. In this episode of Agent Native, host Riley Brown sits down with developer Ross Mike to unpack what the platform consolidation means for technical users and mainstream consumers alike.
The conversation covers the practical differences between the three model tiers: Soul targets power users and developers running long agentic tasks, while Terra and Luna are designed for everyday knowledge work. Ross shares firsthand impressions of GPT-5.6 Soul, highlighting improvements in token efficiency, tool-calling precision, and autonomous task completion compared to GPT-5.5 — including the model’s reduced tendency to pause mid-plan and ask for permission to continue. He also draws a direct comparison to Gemini’s tendency to spiral into excessive tool calls, framing token efficiency as a key differentiator for agentic workloads.
Riley brings a marketer’s perspective, walking through real-world use cases such as using Codex in self-scoring evaluation loops to iteratively improve ad scripts against high-performing competitor examples. The duo also explores OpenAI’s new Sites feature, the updated in-app browser, and the strategic reasoning behind surfacing the “Work” tab to ChatGPT’s 983 million users — a move aimed at closing the perception gap between ChatGPT and more agentic platforms like Claude.
📺 Source: Riley Brown · Published July 12, 2026
🏷️ Format: Podcast
