How We Recommend AI Models For Writers
- Reading time
- 3 min read
- Written by
- by Kate
Every few weeks, a new model arrives, and our users ask us: when are you adding it to the default collections?
A new release often comes with impressive claims, early reviews, and examples of what it can do. At Novelcrafter, we support the release of each new model, but we don’t add it to our default model list immediately.
That doesn’t mean you have to wait to use it. You can access a model as soon as your connected AI vendor adds it, and add it to your personal model collections.
In this blog post, I’m going to go into more detail about what is behind our decisions.
The Model List is a Recommendation
When a model appears in one of our default collections, we’re saying that we think it’s a strong choice for that task. Most writers are looking for a useful balance of price, speed, reliability, and consistency.
There are a few things we consider:
Latest models may not always be popular, or even the best. Recent advancements to make models more agentic, tend to actually hurt their creativity and writing style.
Some models may write well, but are terrible at context recall and handling details. Those may get be used for text replacement instead.
Recent models also tend to think a lot. While this helps with some tasks, we look for models that can get the job done fast and quick enough.
Not all models are served with data privacy and security in mind. Zero data retention (ZDR) of your prompts is important to us, so even if a model checks all the other boxes, this could be a deal-breaker.
Newly released models may have limited capacity while its vendor works to meet demand. It might respond slowly, fail during busy periods, or return errors often enough to interrupt your writing (let me say, it’s no fun when that happens!).
A model’s behavior can also change after release. Vendors may adjust the model, fix problems, change its price, or replace an early version. Waiting gives us time to test it more thoroughly and see how it performs during normal use, not just under ideal conditions.
There are a lot of models
Between them, AI vendors offer access to hundreds (if not thousands) of models.
If we added every new arrival, the default collections would quickly become overwhelming. You’d scroll past dozens of unfamiliar names to reach the one you wanted, and someone opening Novelcrafter for the first time would have no idea where to begin. A shorter collection gives you a useful starting point.
We also try not to update the default collections too often. Imagine you’ve found a model that works well, and have created the perfect prompts for it… If we were to replace that model every month just because a newer version appeared would disrupt your flow, and make your work less predictable.
Why are there Different Lists?
You won’t see the same models in each collection, as we curate each model collection separately. For example, a model that generates unremarkable prose may give succinct and very accurate summaries.
Broadly, this is what we look for:
- Writing prose (scene beat completion): The model should follow your voice rather than fighting it. Dialogue should sound like people speaking, characters should behave consistently, and the writing should stay in the correct point of view.
- Text Replacement: The model should change what you asked it to change and leave the rest of the text alone. Strong instruction following matters more here than producing elaborate prose.
- Summarization: The model should accurately and concisely capture the key events in a scene without exceeding the requested length. Where possible, we prefer a lower-cost model.
- Chat: The model should follow instructions, analyze the context you provide, and act as a writing partner. It should be able to discuss ideas without taking control of the story.
Not on the list? You can use it anyway
Because you bring your own AI connections, you can access any model your vendor offers, whether we’ve tested it or not. You can also build personal collections to keep your favorites together and switch between them as you work.
If you’ve found something that writes beautifully and isn’t in one of our default collections, we’d love to hear about it. User recommendations often help us decide what to test next, although a model still needs to perform well across our wider checks before we add it.
Managing Your Model collections
Kate
Based in the UK, Kate has been writing since she was young, driven by a burning need to get the vivid tales in her head down on paper… or the computer screen.