A developer reports using Anthropic's Fable model via API for under a dollar to generate project designs, then passing those designs to lower-cost models for implementation, achieving results comparable to letting Fable complete the work.
Background
- Anthropic Fable is a rumored or internal high-performance AI model from Anthropic (the company behind Claude), believed to be stronger than the publicly available Claude models. It is not generally accessible via the standard API — the user here likely has special access or is referring to a specific endpoint.
- The post describes a cost-saving workflow: use Fable (on a low setting, "low" probably meaning low token generation budget or low temperature) to generate the design/architecture, then hand that plan to a cheaper, weaker model to write the actual code. The result is nearly as good as letting Fable finish the entire task itself, but much cheaper.
Anthropic announced that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5, and that they will begin restoring access tomorrow with a further update to follow.
Paul Graham speculates about future AI progress, imagining how advanced models might become in five years if they improve upon the current model Fable as much as Fable has improved upon GPT-3. The post quotes Jared Friedman praising Fable as "insanely good" and deserving of its hype.
Peter Gostev spent extensive time testing difficult 3D prompts at Fable and compiled a 45-minute video featuring over 60 demos, with prompts to be shared in a follow-up post.