Anthropic launches new Claude Fable and Mythos 5.2 models

🧠 Claude now "dreams" between chats — so it can fix its own bad notes Anthropic's new engineering post says agent memory works a lot like sleep: Claude writes notes during a session, then offline "dreaming" replays the transcript, corrects mistakes, builds wiki-style maps, and promotes what the next session should load first. In a Pokémon Red test, Sonnet 3.7 dumped 413 memory files and earned one badge. Fable 5 kept nine files and earned four. When Claude wrongly marked Mt. Moon as finished, dreaming checked the transcript, rewrote the note, and Claude went back into the cave instead of looping. On ~350,000 internal company files, scaling dreaming from 2 to 128 sessions lifted a 25-question eval from 13% to 48%. Anthropic's write-up frames dreaming as idle compute that organizes memory — not a change to the model's weights. @aipost 🏴

It is an extreme productivity drain to have these models substitute settled vocabulary with random synonyms, particularly when working on math or code. No, Claude, you cannot just substitute "morph the routine" with "reshape the routine" because only Morph is in the code while reshape could literally be anything. This semantic substitution makes it extremely tedious to understand its work and output and, moreover, occasionally cause it to write a wrong plan that gets interpreted in a different way by the next agent picking up the plan. I almost suspect that this is a result of needing to satisfy watermarking regulation. Now, I'm not saying that watermarking itself caused this. I am instead hypothesizing that as part of a project to satisfy regulation, Anthropic created more semantic substi