The AI writing leaderboard

Top 5 writing models – overall score

What to use, when

Best for sales and sensitive customer-facing copy: Claude Sonnet 5.5

Best for generating clickable headlines and titles: Claude Opus 4.6

Best for coming up with names for products and companies: Fable 5.1

Best for writing according to a tone of voice guide: GPT-6 Astra or Claude Sonnet 5.5

Best for writing about dense topics clearly: Fable 5.1 or Claude Opus 5.5

Best for overall writing: Fable 5.1 or Claude Opus 5.5

The leaderboard

Subscribe to our monthly Def AI update (models, features, use cases) and our quarterly roundup (the best models for your tasks)

How it works

Our scoring panel is made up of specialist writers from the Definition language team.

They have been crafting copy and defining tone of voice for brands like Monzo, Specsavers, Zurich and Disney+ for decades.

We test every major LLM across five core business writing capabilities, using eight different tasks:

1) Customer messages and persuasive copy

• Delivering bad news to customers via email
• Product description

2) Headlines and titles

• Article headline generation
• YouTube title generation

3) Naming things

• Product names
• Company names

4) Tone adherence and clarity

• Adapting copy to match specific tone of voice guidelines

5) Clarity

• Rewriting dense text for clarity

Our specialists review every output and grade the model out of 10 for each capability. A 10 means the copy is “client-ready”, perfectly on brief with zero edits needed.

We add new models to the table soon after their release. Benchmark programme led by Tom Pallot.

Definition AI gives you access to the models on this leaderboard and guidance from our specialists on how to use them.

Our language team panel:

• Nick Padmore, head of language
• Hannah Moffatt, creative director
• Nick Banks, language and AI specialist
• Alex Goldstein, creative director
• Ashleigh Thompson, writer
• Matt Boothman, senior writer
• Robert Hart, senior writer