MYTHOS MYTHOS MYTHOS

TL;DR

• 1

  Mythos is out, and Fable is the guarded public version: Berman says the only real difference is guardrails, with Mythos aimed at security work and Fable 5 offered broadly as Anthropic's most capable generally available model.

• 2

  The model feels qualitatively different, not just better on benchmarks: He calls Fable 5 more different than any model he has tested, saying even tiny prompts triggered broad codebase-level exploration and long autonomous runs.

• 3

  Anthropic is charging frontier prices for frontier tasks: At $10 per million input tokens and $50 per million output tokens, Berman argues Fable is too expensive for routine work but worth it for tasks like Stripe's reported one-day migration of a 50 million-line Ruby codebase that would have taken a team two months.

• 4

  Fable's biggest quirks are verbosity, density, and endless confirmation loops: He says the model's explanations were so technical he repeatedly edited his Claude.md file to ask for simpler language, and it kept asking clarifying questions before acting.

• 5

  Workflows mode turns Fable into a parallel agent swarm: In one test, he watched 100-plus sub-agents fan out across a codebase, and in another fluid simulation task it launched 63 agents in parallel while rapidly burning through tokens.

• 6

  Anthropic's safety posture looks tied to strategy as much as risk: Berman suggests the delayed release and new 30-day retention policy help Anthropic defend against jailbreaks and distillation while preserving its lead over rivals and open source, which he now estimates is closer to a year behind.