Every model launch triggers the same question: "Is this the new best one? Should I switch?"
It's the wrong question, and this week's releases prove it. On November 9, Anthropic shipped Claude Fable 5.2 — billed as its most capable model for technical work — at $8 per million input tokens and $40 per million output tokens. Its sibling, Claude Opus 5.5, runs $4 and $20 on the same scale. Same company, same "frontier" tier. One costs exactly twice the other.
So which is "best"? Neither. The right question is: which of my tasks actually need the expensive lane? Here's the fresh evidence, and a 10-minute exercise to answer it.
What hands-on testing actually showed
Claire Vo — founder of ChatPRD and host of the How I AI podcast, and by her own description a longtime Claude skeptic — spent two days building with Fable 5.2 before recording her review. Five results stand out for anyone running a business:
- It turned a tweet and a brand book into an animated marketing video that, in her words, doesn't look like slides. Perception-to-motion work: the model's clear edge.
- It assembled flat-pack furniture from blurry floor photos, with no manual — a geometry test she reports every other model failed. Reading messy real-world images and reasoning about physical space.
- It extended a set of brand sprites into a complete bug-defense game, trailer included. A multi-step build with a finished artifact at the end.
- It handled a real re-architecture of ChatPRD's product intelligence pipeline — engineering work on a system that already existed.
- And it still fell down on web design — her take: don't let it design a website on its own.
That last point is the one most coverage skips. The lesson isn't "Fable 5.2 is better." It's that frontier capability now comes in shapes — and the shapes don't match the price tags.
The routing rule
Route by task shape, not by ranking:
The default lane (Opus 5.5 at $4/$20). Drafting, replies, summaries, analysis, routine fixes — the bulk of daily work that was already fine on last year's models. Anthropic's own Opus 5.5 launch post concedes the point: at these capability levels, benchmark margins have "become a less reliable guide to real-world differences," and the gap to its frontier model is "narrower than these scores suggest." When the vendor tells you the scores exaggerate the difference, believe them.
The specialized lane (Fable 5.2 at $8/$40). Three shapes earned the premium in testing: visual and spatial work (images, PDFs, diagrams, physical layouts — the furniture test), builds that end in a finished artifact (a video, a game, a tool — the tweet-to-video and sprite-to-game results), and re-architecture of working systems where wrong assumptions get expensive (the pipeline rebuild). Anthropic's own positioning agrees: Fable 5.2's headline use cases are coding, finance, the sciences, and design for 3D, hardware, and motion.
The human lane. Brand-critical design stays human-reviewed no matter which model you point at it. The web-design verdict is the pattern: a specialized model raises the floor on visual work without making taste optional.
One trap worth knowing before you route anything: Fable 5.2 includes safeguards that automatically hand flagged cybersecurity and biology queries to less capable models, and Fable use carries 30-day data retention for safety monitoring by default. If your work touches those domains or sensitive data, check the fallback and retention terms before assuming you're getting what you pay for.
The 10-minute routing audit
- List the AI tasks you actually ran last week. Real ones, not aspirational ones.
- Tag each: everyday work, build-a-thing work, or looks-critical work.
- Point everyday work at the cheaper lane, build work at the premium lane, and keep looks-critical work human-reviewed in both.
- Before trusting either lane with a task family, run it on five inputs where you already know the right answer. A lane that gets four of five isn't "80% good" — it's untrustworthy exactly where the money is.
- Write the routing down. Next month another model launches and the "should I switch?" instinct returns; your notes are the antidote.
Honest boundaries
- Prices and names move fast. Fable 5.2 itself launched 20% cheaper than Fable 5.1 ($8/$40 vs. $10/$50). The durable habit is routing by shape; the specific models are this month's answer.
- Fast mode is a separate purchase. Both models offer a fast mode at exactly double the standard price ($16/$80 for Fable 5.2, $8/$40 for Opus 5.5, per Anthropic's pricing). Buy speed when you're waiting on the model; don't pay it as a default.
- Two days of one expert's testing is impressions, not measurement. Treat her results as a hypothesis about task shapes. Your five known-answer inputs are the actual test.
Tonight
Pick the three AI tasks you ran most this week. Ask of each: does this need the $8 lane, or does the $4 lane handle it? If you can't answer from evidence, that's your next test to run — before the next model launch makes the question loud again.
Want the boring half pre-built? The Boring Automation Pack from SMB Scale Up includes plug-and-play automation blueprints plus the standing-instructions worksheet — the checklists and tracking files that keep delegated work inside lines you actually wrote down.
Top comments (0)