Choose which AI writes your book, down to the scene
Last checked against the product November 10, 2026
AI models do not write fiction equally well, and they are not good at the same things. One handles dialogue better, another pacing, another an emotional scene. In Novelmint you choose. Each series has an AI Models tab that lists the models you can build with, scored on our blind fiction benchmark, with their cost beside the score. Pick a lead model to carry your voice, let others take the scenes they suit, and pin a specific model to any single beat when you know what you want.
Key takeaways
- Every series has an AI Models tab. You pick a lead model, which writes most of each chapter and its high points, and can include others for the scenes they suit best.
- Each model shows an overall craft score out of 100 from blind benchmarks, a breakdown by skill, its cost, and how willing it is with mature or explicit content.
- A "More models" list adds many more models through OpenRouter. They can lead a series now and are tested on the Bake-off tab, because our benchmark has not measured them yet.
- Any single beat can be pinned to a model of your choice, or left on Auto. A beat is one story moment, such as a scene or an exchange, and a chapter is a handful of them.
- Set a Studio default for every series, and give any one series models of its own.
What it does
What you can choose
A lead model
The lead writes the bulk of every chapter and owns its high points, so the series keeps one voice.
Supporting models
Tick other models and the build draws on them for the individual beats they suit best, while the chapter still reads as one piece.
Scores you can read
Each model’s overall craft score runs from 0 to 100, from blind benchmarks judged by models from other makers. Open a model for its breakdown: literary quality, dialogue, sticking to the brief, pacing, and content areas such as emotion, conflict and romance.
Cost and content, side by side
Each row shows the model’s cost relative to a fixed baseline, its capability tier, whether it writes mature or explicit content, and whether your plan covers it. Sort by best craft, cheapest, or A to Z.
More models
The "More models" filter searches a much larger list from OpenRouter, from makers such as Mistral, Meta and Qwen. They run on hosts that keep no copy of your writing and are billed in credits at each model’s own price.
One beat, one model
In the prose view, each beat has a model menu. Leave it on Auto, or pin the beat to a model and it is used for that beat at every build and every re-render. Under Auto, a small caption shows which model wrote it.
How to use it
Setting it up
Set your Studio default
Open the AI Models tab in your Studio. The models you choose there are used by every series that has no models of its own.
Tune one series
On a series, open its AI Models tab and change anything. That series now has models of its own. "Use my Studio default" puts it back.
Compare before you choose
Sort by craft or by cost, and open a model to see where it is strong. For a model with no score yet, run it against others on the Bake-off tab.
Pin a beat where it matters
Switch a chapter to prose view and use the model menu on any beat. A pinned beat keeps its model when you build the chapter or re-render that beat.
Or use your own key
If you have an API key with an AI provider, pick a model from it on the same tab. The whole series then builds on that one model.
Questions
Frequently asked
- Can I choose which AI model writes my book in Novelmint?
- Yes. Each series has an AI Models tab where you pick a lead model and, if you like, others the build may draw on for particular scenes. You can also set a default for every series from your Studio, and pin a model to any single beat.
- How do I know which model writes fiction best?
- Each model on the tab carries an overall craft score and a breakdown by skill, from Novelmint’s blind fiction benchmark. The full results are public, and you can run the same test yourself on your shortlist with a bake-off.Fiction model benchmarkModel bake-off
- What is the difference between the lead model and the others?
- The lead writes most of each chapter and its most important moments, which keeps one voice across the series. Models you include alongside it are used for individual beats they suit better, such as a dialogue-heavy exchange.
- Can I use a model for just one scene?
- Yes. In prose view every beat has a model menu. Pin a model and that beat is written by it at every build and re-render; leave it on Auto and the build picks. Under Auto a caption shows which model wrote the beat.What is a story beat?
- What are the "More models"?
- A larger list of models offered through OpenRouter, from many makers. Any of them can lead a series. They are not mixed in as supporting models until the benchmark has measured them, so try them on the Bake-off tab first. They run on hosts that keep no copy of your writing.
- Do different models cost different amounts?
- Yes. The tab shows each model’s cost relative to a baseline, and the build estimate follows the models you pick. Current credit prices and plans are on the pricing page.Pricing
- Can I use a model from my own API key?
- Yes. Add a key from Anthropic, OpenAI, Google, xAI or OpenRouter and pick any model it reaches. While a key model is set, every beat in the series is written by it, and your per-beat picks wait until you switch back.Bring your own AI key
What this page does not claim
- A craft score is a measured average on standard scenes. It does not promise how a model will write your book, which is why the bake-off exists.
- Models from the "More models" list have no score until the benchmark measures them.
- The list of models changes as makers release and retire them. This page names none on purpose; the tab and the benchmark show the current ones.
- Choosing a model does not replace the plan. Every model writes from your beats, your story bible and your voice settings.
Keep reading
Where the scores come from
The craft scores come from a public benchmark: every model writes the same scenes, and judges from other makers score them blind. You can run that same benchmark yourself on the models you are weighing.
Related
Plan a book with Minty
Talk the story through and watch it become chapters and scenes on your Timeline.
Chapter builds
Turn a chapter’s beats into prose, written from your plan, your bible, and your voice.
Bring your own AI key
Build chapters on your own Anthropic, OpenAI, Google, xAI or OpenRouter key, or on the AI your credits include. Your choice, per series.
Model bake-off
Run our fiction benchmark on up to five AI models, ours or ones on your own key, and read their scenes side by side with blind judges’ scores.