Claude's Next-Gen AI: Is it a Paradigm Shift or Pricey Power?

Summary

Anthropic's new Claude Fable 5 and Mythos 5 models are being hailed as a paradigm shift in AI capabilities, with Mythos 5 boasting significantly higher performance in benchmarks like Cursor Bench, achieving a 72.9% score on Max. While Fable 5 is the safeguarded version available to the public, Mythos 5 has its safeguards lifted for select partners. The pricing for Fable 5 Max is $10 per million input tokens and $50 per million output tokens, making it expensive, though less so than initially anticipated. The speaker demonstrates Fable 5's potential by tasking it with building a mobile companion app for a business AI platform called Pluto, and highlights Mythos 5's ability to fix complex bugs in the Pluto codebase, including issues with chat streams and iMessage connections that other models couldn't resolve. The speaker also notes that Mythos 5 has consistently received high marks from code review agents like Gravile. A "rug pull" is anticipated on June 23rd when Fable 5 will be removed from Pro Max and Team plans, requiring usage credits thereafter. Anthropic is reportedly constrained by compute, even paying SpaceX for GPUs. The speaker contrasts the current excitement with past AI skepticism and addresses concerns about pricing, calling it a "rich man's model." An interesting point from Anthropic's system card is that Fable 5 and Mythos 5 share identical weights, with Mythos being the unrestricted version. Mythos 5 is classified as a CB1, not CB2, indicating a less clear judgment than previous models. Estimates suggest Mythos 5 can produce the equivalent of 72 working days of work for two people in 16 hours, and generalist biologists using it outperformed specialists in a tabletop exercise. Furthermore, Fable 5 includes a hidden safeguard against its use for training frontier models.