What is the best AI model for writing?

The model at the top of this board, which people picked most often in blind votes between two drafts. The board updates as people vote, so the top of the table is the leader today.

How are the votes collected?

In the AI writing model test, people judge two drafts written from the same brief and pick the better one. Neither draft shows its model's name or price until the end.

Which writing jobs do the models do?

Ten jobs creators do every week, from a YouTube intro and an X thread to show notes, a sponsor pitch and a launch email.

How is the rating worked out?

With the Bradley-Terry model, the method the big blind-vote leaderboards use. Beating a strong model counts for more than beating a weak one.

What does the 95% interval mean?

It's the range the model's true rating most likely sits in. When two models' ranges overlap, treat them as tied, whatever their ranks say.

Why is a model missing from the board?

A model shows once enough votes include it for its rating to mean something. Until then its interval would be too wide to rank it.

Yes: take the AI writing model test, and every choice you make counts as a vote.

Are the votes anonymous?

Yes: a vote stores which draft won and which lost, with the browser's random id and the IP address hashed, never a name or an email.

Are the drafts real AI outputs?

Yes: every model wrote its drafts from the same brief through OpenRouter, and nothing was edited.

Which models are on the board?

The newest model from each major AI company as of September 30, 2026, plus a cheaper sibling where there is one.

Can I download the data?

Yes: the board's CSV lists every model with its rating, its interval and its votes.

Best AI model for writing

Find the best AI model for writing, ranked by people's blind votes between two drafts written from the same brief, with the names and prices hidden.

Updated Oct 1, 2026

Your votes build this board

The board fills as people vote: a model shows once 20 votes include it. Judge a few drafts to help it fill.

Take the blind test

No model has a score here yet.

Source: people's blind votes between two drafts written from the same brief, in the AI writing model test on navid.me

Every leaderboard

Each board has its own page with every model, a chart and a CSV.

LLM leaderboard

Claude Opus 5.5 leads, from Anthropic

Best AI model for coding

Claude Fable 5.1 leads, from Anthropic

Best AI model for agents

Claude Fable 5.1 leads, from Anthropic

AI model pricing

Cheapest now: Mistral Nemo

Best AI model for website design

Muse Spark 1.3 leads, from Meta

Best AI model for game development

Claude Opus 5.5 leads, from Anthropic

Best AI model for UI design

Claude Opus 5.5 leads, from Anthropic

Best AI model for data visualization

Claude Opus 5.5 leads, from Anthropic

Best AI model for SVG graphics

Claude Opus 5.5 leads, from Anthropic

Best AI model for 3D scenes

Claude Opus 5.5 leads, from Anthropic

Best AI model for building mobile apps

Claude Opus 5 leads, from Anthropic

Best AI model for building full-stack apps

Claude Fable 5.1 leads, from Anthropic

Best AI image model

GPT Image 2.5 Sunburst (max) leads, from OpenAI

Best AI image editing model

GPT Image 2.5 Sunburst (max) leads, from OpenAI

Best AI video model

Wan 3.0 leads, from Alibaba

Best image to video AI model

Gemini Omni Flash leads, from Google

Best AI voice model

Eleven v4 leads, from ElevenLabs