CharlieKey takeaways from your favorite podcasts
← 11 November 2026The AI Daily Brief

Fable 5.2 vs Opus 5.5 and When to Use Each

Fable 5.2 is the most powerful model yet, excelling at big, ambitious, open-ended projects, but it is jagged, expensive and often no better than Opus 5.5 for everyday work.

Watch on YouTube ↗

Charlie's summary

  1. Fable 5.2 lifts the bar across benchmarks, and its Remote Labor Index score nearly doubled Opus 5.5's

    Fable 5.2 leads on coding, agentic science and the Artificial Analysis index (65, six points above Opus 5.5). On the Remote Labor Index, which uses real contractor tasks, it scored 41.3% against Opus 5.5's 21%. The host quotes Garrison Lovely, who calls it the most alarming score: the best agent managed just 2.5% a year ago, a reminder of “how fast change can come during exponential growth.” Knowledge-work gains were smaller.

    ▶ 2:30
  2. Fable 5.2 is cheaper per token but very costly per task, so effort settings decide the economics

    Anthropic cut per-token prices 20%, but on max effort a task costs $26.54, 2.6 times Opus 5.5. Dropping to high brings it to $9.08, about 10% below Opus 5.5 on max. Anthropic says medium should be the default, high is enough for everyday coding, and max is for finding the ceiling. Matt Schumer thinks even low suffices for most tasks, so real costs may differ from the headline numbers.

    ▶ 5:03
  3. Anthropic is reserving Fable for ambitious work that other models can't handle

    The showcase examples are huge. Over eight days, 1,000+ sub-agents built a working operating system from an empty folder. The model formalized Fermat's Last Theorem in 30 hours, where Opus 5.5 got 26% of the way. It also solved two open black-hole physics problems and modeled 28 of 34 capsids. The host reads this as guidance for the rest of us: use Fable where other models stall.

    ▶ 6:37
  4. Early testers say Fable 5.2 can build production-grade software end to end, reviving the build-your-own-software question

    The Every team says Fable 5.2 can take a whole product and keep building, testing and refining until it is something you use daily, letting professionals build tools without reading the code. Their example was a video editor combining Premiere-, After Effects- and Ableton-style capabilities. The host notes this reopens the SaaS-apocalypse debate, since maintenance burden was previously the argument against building your own tools. Testers also report strong security audits: Pavel Huryn's bug-hunt benchmark found 58 of 105 planted bugs.

    ▶ 13:20

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.