Fable

Fable is the model ToTheMoon has talked about most in the summer of 2026: from episode 121 on it turns up in nearly every conversation about the hosts' own projects. They treat it less as news than as a daily working tool, and they ask two questions at once — what the model changes in long engineering work, and what it means to depend on a tool to which access can be cut off like a strategic resource.

The first impression in episode 121 was about stamina: the model holds a long sequence of actions and gives up less often where earlier versions suggested switching to Codex. Episode 124 recorded the access shutdown and noted that Fable was forcing OpenAI, Google and Meta to prepare answers. When access came back (episode 130) the hosts talked about dependency risk — a team could move a project onto the model and lose it — and recommended splitting work between Fable, Codex and a cheap model. The extension of access in episode 133 was read as a reaction to OpenAI's releases.

By late summer the verdict was practical. In episode 154, discussing why new versions do not always beat old ones, the hosts named Fable the exception: it brought a different quality of result, as if an adult had taken over a task from a teenager. Episode 153 describes a system built entirely by agentic Fable together with ChatGPT Pro and Claude Design, and in episode 156 Fable is called the main system. Episode 150 added a gripe: Claude sometimes switches from Fable to Opus when it decides a task is a security gray area. Episode 161 covered Fable 5.1 as an update built for large systems rather than for chat. In episode 162 Fable became the yardstick for Astra: by OpenAI's account the new GPT-6 model substantially surpassed it, and the host compared the two not as chats or agents but as a space you work in.