Hemmingway vs Sudowrite

Sudowrite is an AI writing tool made for fiction writers. It is built around the work of writing a novel or a long story: drafting scenes, describing, brainstorming and revising, inside a place that knows it is a book. Hemmingway is a general writing tool built for writing that sounds like a person: messages, emails, letters, and short pieces such as a story, a poem or a toast.

If you are writing a novel, the honest answer is that Hemmingway is not the tool to pick. In our own StoryBench test, the largest story models write better long fiction than our model, Hemmingway-1. Sudowrite is built for that job. Hemmingway suits short creative pieces and the everyday writing around them.

The short version

We did not test Sudowrite. StoryBench compares models, not writing apps, so there is no Sudowrite score on this page.

What StoryBench shows

StoryBench is our own story test. Each model writes stories to the same prompts, and every story goes head to head with another model's in a blind judgement, run in both orders so position cannot sway the result. The judge is never one of the models being judged. Scores are on an Elo scale: higher wins more often.

Fable 5 Max and GLM-5.3 write better stories than Hemmingway-1 in this test, and so do the newest frontier models. Hemmingway-1 sits level with Kimi K3, ahead of Qwen3.8 Max and DeepSeek V4 Pro, and 504 points above the model it was built from. For a 27B model that is a lot. It is not first.

When we broke the results down by kind of request, Hemmingway-1 lost on long story turns and on hostile storytelling. The big story models are better at those, and that is fair. More on AI for creative writing.

Where Hemmingway does fit

Fiction is not only novels. Plenty of creative writing is short, and has to sound as if someone sat down and wrote it: a story for a child, a poem for a card, a wedding speech, a letter in a character's voice, a piece of flash fiction.

That is where Hemmingway-1 is strongest. In our Human-Likeness test, where a judge picks which of two texts a person wrote, it came first of the models we tested, twenty-six points clear of the next. On EQ-Bench 4, a public emotional-intelligence benchmark we ran with its official harness, it scored 1330, third, behind Fable 5 (1341) and Kimi K3 (1332). Getting a character's feelings right is a large part of short fiction.

Here is a story Hemmingway-1 wrote, unedited, for the prompt "Write a story of about 150 words about a lighthouse keeper who gets a letter meant for someone else."

The supply boat came up the rock in November rain, and the boy who brought the stores had one more thing: a letter, addressed to Mr. Brannan, keeper of Vane Rock Light.

Callum took it without opening it, as he always did. After the lamp was lit and the log signed, he broke the seal. It was from a woman in Oban, Brannan's sister. She said she was leaving the mainland in spring. She said the things she had not said in the autumn were said now, all of them, and that she did not expect an answer.

Callum had kept letters like this for thirty years, for other men, other lights. None had ever been sent on. He read hers twice, then folded it along its old creases.

At four bells he climbed the stairs and trimmed the wick. The beam went out over the water, and the letter stayed in the drawer.

It kept to the length it was asked for and ended without explaining itself. That is the kind of piece it is built for.

Getting good short fiction out of any AI

These steps work in Hemmingway and in most tools.

1. Give it a length. "About 150 words" gets you a story you can read in a minute. "A short story" gets you whatever the model thinks short means.
2. Give it one concrete situation. A lighthouse keeper with the wrong letter is a story. "A story about loneliness" is an essay waiting to happen.
3. Name what to leave out. "No moral at the end, no title, don't explain the ending." Models like to tie things up.
4. Say who it is for. A bedtime story for a six-year-old and a piece for a literary magazine need different words, even with the same plot.
5. Ask for one version. Then ask for changes to that one, rather than choosing between three.
6. Read it aloud and cut. Take out any line you would not have written. It is your piece.

Try HemmingwayDownload the app

If you are writing a novel

A novel is a different job. It needs a model that holds a long plot together, keeps characters consistent across chapters, and writes well at length. A tool made for novelists, such as Sudowrite, is built around that work, and the big story models are better at long fiction than Hemmingway-1 in our tests.

Hemmingway-1 does have a 262,144-token context, so it can read a long draft, and a reply can run to 32,768 tokens. But being able to read a long manuscript is not the same as writing the best chapters, and on long stories our own test says it is not the leader.

There are still smaller jobs around a novel where it can help: the query letter to an agent, the short blurb, the email to your editor, a scene that needs to sound like a real text message between two characters. Those are writing to people, which is what it was built for.

Open weights

Hemmingway-1 is open weights under Apache-2.0, including commercial use. You can download it from Hugging Face, run it yourself or fine-tune it, or use it through an OpenAI-compatible API as hemmingway-27b (docs). If you want to build your own writing tool, that is something you can do with it. More on open models for writing.

Privacy

We never train our models on your chats, memory or People. For anyone writing something they have not shown anyone yet, that matters. Our privacy page lists everything we keep.

Common questions

Is Hemmingway good for writing a novel?

Not as good as the largest story models. In our own StoryBench test, Fable 5 Max and GLM-5.3 beat Hemmingway-1 on stories, and so do the newest frontier models. For a novel, use a tool built for fiction or one of the big story models.

What is the best AI for short stories?

For short pieces that need to sound like a person wrote them, Hemmingway-1 is strong: it came first on our own Human-Likeness test and third on the public EQ-Bench 4. For long fiction, the big story models lead. Give any model a length, one concrete situation and a note to skip the moral, and you will get a better story.

Did you test Sudowrite?

No. StoryBench compares models, not writing apps, and Sudowrite was not in it. This page compares what each is built for, and gives our model's scores against the models we did test.

Will Hemmingway use my manuscript for training?

No. We never train our models on your chats, memory or People. The privacy page lists everything we keep.