Summary
AI educator Julia McCoy reviews OpenAI’s GPT 5.6 Sol inside the newly merged Codex workspace, testing its ability to grind through long, multi-step tasks without stopping. She breaks down OpenAI’s new naming system—Sol as the flagship model, Terra as the balanced mid-tier, and Luna as the fast, cheap option—along with their API pricing ($5/$30 per million tokens for Sol, cheaper rates for Terra and Luna) and the new effort slider ranging from instant to extra-high reasoning.
The video covers Codex’s expanded feature set, including in-thread result visualization, built-in app hosting and database deployment, a functional mobile app for steering tasks remotely, and “Ultra mode,” which runs multiple Sol sub-agents in parallel on the same job. McCoy cites the Agents Last Exam benchmark, where Sol scored 53.6, the highest reported at the time, as evidence for its persistence on long-running tasks.
As a longtime Claude user, McCoy offers a balanced comparison, concluding that Sol excels at follow-through and task completion while Claude’s Fable 5 retains an edge in architectural judgment and planning. She also flags a practical gotcha: ChatGPT’s consumer version of Sol and the version running in Codex diverge after an August update, which can cause inconsistent behavior.
📺 Source: Julia McCoy · Published September 28, 2026
🏷️ Format: Review
