GPT-5

On ToTheMoon, GPT-5 is a story about the gap between expectation and a working product. The hosts followed the model from Sam Altman's first promises through versions 5.2, 5.4 and 5.6, and each time the question was not how much smarter it had become but what had changed in their own work, and why a stronger model keeps forcing users to sort out its modes.

In February 2025 OpenAI said it would skip a standalone o3 and concentrate on a GPT-5 that decides for itself whether to answer with voice, an image or a deep search (episode 046). Claims of GPT-5 and AGI within two months the hosts treated as hype (episode 054). Live tests after the release showed a faster, more assured dialogue, but hallucinations were less visible rather than gone (episode 070). A week later the summary was soberer: the model was not head and shoulders better, the release seemed aimed at free users, and choosing between Thinking, Fast and Pro had become almost harder (episode 071).

From there the channel tracked the versions as a stream. GPT-5.2 arrived so quickly that the only useful question was what had changed at work, since users pay for a stable tool rather than a loud launch (episode 089); in 2026 the hosts also noted that physicists had published discoveries co-authored with GPT-5.2 Pro (episode 098). GPT-5.4 makes clear progress on hard tasks, yet the interface again asks the user to choose between Instant, Thinking and Pro (episode 101). The Sol, Terra and Luna lineup for GPT-5.6 got the same verdict: the split may simplify the economics, but it adds more names to understand (episode 130).