If you already pay for an AI key, bring it to Novelmint!

Ben, Founder of Novelmint.ai here. I have an exciting new feature to tell you about!

Your votes on our public roadmap led directly to this feature. Thank you for asking for it! Your suggestions and support help us grow and improve our product, and it means a lot to see you all so active and engaged on Novelmint. Thank you!

BYOK - Bring Your Own Key

If you already pay Anthropic, OpenAI, Google, xAI or OpenRouter, you can add your key and build chapters on any model it reaches within Novelmint - and Medium builds from Novelmint won't charge you to do so. It was the #2 idea on our public roadmap, and I understand why: a lot of you already pay for a model, and paying a second time to use it here never sat right with us either.

Your keys go under Account → AI keys, stored with full encryption. Then on a series' AI Models tab you pick ONE model from your keys, and that model plans, writes and cleans up the chapter on your key, so you pay your provider at your provider's rates. If you build at High or Max, we add our judging and rewrite passes on top, and those are charged in credits because they run on our side. The key pays for chapter builds only - Minty, audio, translations and covers stay on Novelmint. Though in the future, you may be able to do those as well.

If your key runs out of funds or stops working during a build, the build stops and tells you, and you press Build again once it's sorted. It will never quietly switch to one of our models and spend credits you didn't choose to spend. And your free first chapter isn't used up by a key build - it's still there if you want it.

150 new Models to choose

BYOK raises the obvious next question - which model should you actually use? That question got even bigger this week because 150 more models from OpenRouter are now available for your chapter builds. From makers like Mistral, Meta and Qwen, and many others, all in one searchable list. Novelmint uses each model's own pricing, so switching models can affect the credits you spend. They run on hosts set to keep no copy of your writing and never train on it, though a host may hold some data briefly where its safety rules require it. If you're on Unlimited, they're covered the same way our own models are: Medium is included, and High and Max pay the extra. You can also set your models once in the Studio as a default and have that choice reflected in every book you work on.

Benchmarking is now in your control

So how do you choose? Typical benchmarking processes tell you almost nothing about whether a model is any good at writing. That's why we built our own benchmarking system at novelmint.ai/benchmarks. Every model writes the same briefed scenes, and an AI judge panel scores each passage blind, 0 to 100, on nine axes: literary, dialogue, fidelity and pacing for craft, then emotion, physicality, conflict, romance and eroticism for the kinds of scene a book is made of.

Those 150 new models we added have NOT been through the benchmark yet. And the reason isn't cost. New models appear all the time, and we wanted you to remain in control over what models you used. That meant giving you the ability to run our internal benchmarking system yourself. We call it the Bake-off.

The Bake-off

The Bake-off is our benchmark, run by you. It's a tab on the AI Models page, and anyone signed in can use it. Pick 2 to 5 models - ours, ones on your key, or anything from the new list - and each one writes our standard benchmark scenes, from 3 to 9 at a time. The same judges score them blind. Then you read the passages side by side with the scores next to them, and you can hide the names until you've made up your mind, which is honestly the best part. You can download the passages as a file, and one click on "Use for Builds" makes your pick the model you write with. Runs cost a small amount of credits and you see the price first so it's not a surprise.

They're our scenes, not your chapter, and that's on purpose: it keeps every run comparable with every other run. One thing to keep in mind is that a run only compares the models you put in it, so treat the score as a good signal and trust your own reading of the passages. You're the one who has to live with that voice for 80,000 words.

Your scores also feed the public benchmark. Once a model has enough scores on every craft axis, it shows up on /benchmarks under "More models, measured by authors." No single author can put a model on the board. Together we can measure a lot more than any one company would.

If you run a Bake-off, tell me what surprised you. I'd love to hear which model beat your expectations.

As always, Thank You for being part of the Novelmint community! We love seeing what you all create!!

Ben