Claude Opus

On ToTheMoon, Claude Opus is the working model the hosts build their own projects with and the reference point everything else is measured against: Fable, GPT-5.5 and 5.6, Grok, Gemini. The question is not whether it is strong but how a user should choose between versions and modes when tokens vanish in minutes and the model race becomes hard to follow (episode 130).

The first mention, in May 2025, is Opus 4 with record results — stronger than o3 and Gemini in code, but noticeably more expensive (episode 059). In 2026 the conversation centers on Opus 4.8. In episode 124 the hosts suggest export controls may deem Opus 4.8 and 4.7 significant and Anthropic will close access. One host built his site with ChatGPT 5.5 Pro and Opus 4.8 Max (episode 135), noted that Opus 4.8 inside Claude Design performs much worse than inside Claude Code (episode 136), yet found that there it still beat GPT-5.6 (episode 140). In episode 130 Opus 4.8 sits among the models that eat tokens fastest.

Opus 5 was met without enthusiasm. Episode 143 argues that the more autonomous Opus 5 becomes, the more important it is to understand why it chose an action and how to stop a faulty chain, and in episode 145, a few days after release, the host admits that in Ultra modes the system works worse for him than 4.8 Max did. Episode 153, on Anthropic's experiments, reports that with eighty agents Opus 4.6 opened nine hundred and eighty change requests, most of them in conflict, while Opus 4.8 reduced conflicts by isolating agents and scored eighteen percent on the shared-fact task against eighty-five for Mythos.