Matt Wolfe tries Anthropic’s newly released Fable 5.2 after a few hours of access. The headline’s “world’s best” is his interpretation of launch benchmarks, not a ranking established by his two hands-on tests.

What the release claims

  • Anthropic says Fable and restricted-access Mythos share an underlying model; Fable adds misuse safeguards. The company recommends Opus 5.5 for most everyday work and Fable for harder technical tasks.
  • Anthropic lists Fable at $8 per million input tokens and $40 per million output tokens. Its benchmark comparisons use different effort settings; do not translate a benchmark task’s cost into a general user-task bill.
  • Wolfe’s Gary Busey SVG test took about 19 minutes 50 seconds and nearly $5 on his setup. His model-judged ranking is a personal experiment, not a standardized evaluation.

The two creator-run tasks

  • At medium reasoning, Fable produced a browser-playable horde-survival game in about 16 minutes. Wolfe tried movement, sprinting, stamina and level-ups, and liked the mechanics more than the visuals or audio.
  • He prompted a handheld button-driven pixel-art gadget. The model returned a parts list, wiring instructions, firmware files, printable case files and, on a second high-effort pass, an illustrated assembly manual.
  • Wolfe opened the generated files in a slicer but did not buy parts, print the case, wire the circuit or verify the device works. This is a design-package demo, not a built-hardware result.

Separate demos from evidence

  • Anthropic’s operating-system and physical-device examples are company-selected demonstrations. Wolfe also shows third-party social-media game, video and app clips, which he did not reproduce.
  • One community “mirror Earth” demo he tried visibly glitched and had frozen figures. Screenshots and short play sessions cannot establish full-game reliability.
  • Use a real task, record effort setting and total spend, and compare the finished artifact with an everyday model before paying for a more expensive one.

“I’ve only been playing with it for a couple hours since it’s come out.” — Matt Wolfe