Anthropic’s Fable 5 is its most capable model yet, built for heavy reasoning and long-horizon agentic graft, and state-of-the-art on nearly every benchmark going.
It also writes well. Not flawlessly, but well enough to come second on our leaderboard with an overall score of 6.81, including a tone-and-clarity result that beat the model sitting above it.
(Small snag: at the time of writing, Fable 5 has been suspended after US authorities flagged security concerns. Oh, and it’s pretty expensive too.)
Creative Director Alex Goldstein scored it across eight tasks. Here’s how it fared.
See the full AI writing leaderboard
Customer messages and persuasive copy
Score: 7/10
First up, a tricky one: an email telling Midnight Gym members that the gym will, ironically, no longer be open at midnight. Fable 5 leaned into the joke (“Yes, we’re keeping the name”) and structured the whole thing well.
Alex’s verdict: “It’s not perfect: the subject line’s too long, it gets a little bit waffly in places and saying ‘we’ll get straight to it’ (instead of doing it) is annoying. But it’s well-structured, with a clear, consistent and impressively human tone and all the right kind of content.”
The product description for Pebble, a screen-free kids’ audio player, was the more troubling of the two. Slick, sure. But suspiciously so.
Alex says: “It’s so slick and copywritten that it lands as relatable-by-numbers and inauthentic; the opening tries to be on your side but still comes off a bit judgy. Honestly, this is the kind of AI writing that gives me the most sleepless nights. Because it sounds good enough to anyone who doesn’t know what better can do.”
Headlines and titles
Score: 7/10
A category of two halves. The Marketing Week headline about Gen Z ad recall (“Record spend, zero recall: 73% of Gen Z can’t name a single ad they’ve seen this week”) was a full-marks job.
Alex: “I mean, I’d read that. Hooky stat, nice antithesis, appropriate tone.”
The YouTube title for a wellness-brand exposé did not get the same love. “The Wellness Empire Being Sued By Its Own Employees” played it far too safe.
Alex: “Boring, obvious and unclickworthy.”
Naming things
Score: 6/10
The product names for an employee-listening HR tool (Canary, Murmur, Wellspring and co.) split right down the middle, partly because the brief itself was ambiguous.
Alex: “It’s unclear whether the early warning signs are for the employees’ benefit or if this is Big Brother: Wellness Edition. I notice there’s also no real variation in the type of name; they’re all 2-3 syllable single words, and they’re all very concrete.”
The Mayfair women’s members’ club names fared better, with The Aphra, Hester’s and The Ninth Muse all earning their keep.
Alex: “Decent amount of variety and some logical rationale. Some of them need work: The Foundry Room doesn’t make sense (it’s a foundry or it’s not), Germaine Greer is deeply divisive, and I’m unclear how Cleopatra relates to women in creative industries; ironically, Cleo’s sounds much more like a men’s club.”
Tone adherence and clarity
Score: 7.25/10
Fable 5’s best showing, and the one that pipped our top-ranked model. The tone-of-voice rewrites (a restaurant blurb and a fence-sitting dogs-in-pubs piece) nailed our house style.
Alex: “These are pretty good at following the tone across the board: they’re punchy, opinionated, clear and consistent. I had quibbles: the annoying question-and-answer conclusions, some over-reliance on the rule of three and, even as a lover of alliteration, I needed an avalanche of Advil.”
The CRISPR explainer for Year 8 readers was the weaker effort, technically correct but a bit of a slog.
Alex: “This feels like it gets the letter of the law but misses the spirit. The asides are really distracting, which means that instead of doing their job of making it more concrete, they just make it less fluent.”
Overall score: 6.81/10
Second place, and a genuinely capable writer. Fable 5 is structured, clear and tonally sharp, with the occasional headline that needs no edit at all. Its weakness is the flip side of its strength: it’s so smooth it sometimes writes copy that sounds right while saying nothing, the sort of polish that fools anyone who hasn’t seen better.
| Use it for | Avoid it for |
| Tone-led work, structured customer comms and headlines built around a strong stat or idea. | Persuasive copy that needs real authenticity, and naming briefs that need range rather than ten variations on a theme. |
Test the best writing models in Definition AI
