Writing & marketing

Let’s open with the house rule, because it shapes the whole podium: an OpenAI model should never see the podium in this category. OpenAI models are complete ass at writing. That’s not contrarianism - it’s the single most upvoted position in the relevant community discourse, where the definitive thread’s top comment reads ‘ChatGPT just talks like it’s posting to LinkedIn at this point’ (544 points), followed by users reporting that no amount of custom instruction fixes it. Claude models are significantly better here across the board, and the gap is structural: writing quality tracks model size, and Anthropic ships the biggest models.

Why Fable 5 takes gold

The largest model in the world is the best writer in the world, and every measure we trust agrees: the highest human-preference Elo of any tracked model, a clean sweep of all eight genres in the only blind human-ranked head-to-head (vendor-published - direction credible, margin unverified), and third-party consensus on the things marketers actually need - voice control, register, long-form coherence, complex rewrites. If your job is producing words a human will read, this is your daily driver.

One caveat we print every time: the refusal tax. Fable’s safety classifier has flagged the word ‘cancer’ as a biosecurity risk, on the record, from a named immunologist. If you’re writing marketing for health, biotech or security, budget for friction - and yes, the irony of the best writer being the one that occasionally refuses to write is not lost on anyone.

The runners-up

Claude Opus 5 inherits the family trait at a fraction of the price. Honest disclosure: nobody has published a writing eval of it yet - the silver rests on model-class inference, and we’ll firm it up or walk it back when real evidence lands.

Gemini 3.1 Pro is the underrated pick, and this bronze is a flag-plant: it gives off a very clear big-model smell in prose, and almost nobody is talking about it as a writer because the discourse only measures Google models on code. Take it for a spin on long-form before you dismiss this.

What about Kimi K3? It tops the biggest creative-writing leaderboard by a huge margin - judged by a Claude model, so the bias runs against it, which makes the result hard to dismiss. It misses the podium because this slot is writing and marketing: voice control, brand register, positioning - all untested for K3, and its documented do-exactly-what-you-said literalism is the wrong temperament for judgement-heavy copy. Creative fiction? Genuinely worth a look.

What to do now

Non-technical teammates picking one model: Fable, or Opus on a budget - paired with coreyhaines31/marketingskills, still the best skills collection for marketers and salespeople going. And whatever you do, don’t let the LinkedIn voice near your brand.