You already pay for Claude or ChatGPT. Both write well.
Now there is Hemmingway-1, a 27B open-weight model that does nothing but write. Its makers say it beats both at sounding like a person.
Here is the short version. For short messages that go out under your name, their tests put it level with Claude and ahead of ChatGPT. For research, structure and long fiction, the big models are stronger, and Hemmingway’s makers say so themselves. Keep the assistant you pay for, and add Hemmingway-1 for the messages.
I have not run the comparison myself yet. The numbers below are the makers’ own, checked against the public board where there is one. The five prompts further down let you check them without taking anyone’s word, mine included.
Disclosure: I build StashBase, which runs Claude Code and Codex on your files, can rewrite a selected passage with Hemmingway-1, and hosts a free Hemmingway-1 demo.
What do Hemmingway’s tests show?
A tie with Claude on short messages, a clear lead over ChatGPT, and a loss on long stories. These are the makers’ numbers from the model card.
| Test | Hemmingway-1 | Claude | ChatGPT |
|---|---|---|---|
| Everyday messages | 1026 | Fable 5.1 1024 | GPT-6 Astra 976 |
| Sounds like a person | 1032 | Fable 5.1 1006 | GPT-6 Astra 964 |
| EQ-Bench 4, public, their run | 1330 | Fable 5 1341 | GPT-5.5 1316 |
| Long stories | 1197 | Fable 5 Max 1277 | ”Behind GPT-6 Astra” |
Is Hemmingway-1 better than Claude for writing?
On short messages it is a tie. On anything longer, no.
(1) Everyday messages: level.
1026 against 1024. The makers call Hemmingway-1 and Fable 5.1 “level within the margin.”
(2) Sounding like a person: a 26-point lead.
This is the test where the judge was asked which of two replies a person wrote.
(3) Claude wraps the message.
The makers report that Fable 5 put its answer inside commentary, options or notes in more than nine replies out of ten. Hemmingway-1 returns only the message. This one is easy to see for yourself with the first prompt below.
(4) Long stories and EQ-Bench go to Claude.
Fable 5 Max beat it by 80 points on long stories, and Fable 5 is ahead on EQ-Bench even in the makers’ own run.
Is Hemmingway-1 better than ChatGPT for writing?
On the makers’ tests, yes, by a wider margin than over Claude.
(5) Everyday messages and sounding like a person: 50 and 68 points ahead.
1026 against 976, and 1032 against 964, both over GPT-6 Astra.
(6) Hard asks show the biggest gap.
On messages like turning someone down, the judge took Hemmingway-1’s reply for the human one 72% of the time, and GPT-6 Astra’s 9%.
(7) EQ-Bench is close, and long stories go to ChatGPT.
GPT-5.5 scored 1316 on EQ-Bench, 14 points below Hemmingway-1. On long stories the makers put Hemmingway-1 behind GPT-6 Astra.
What do the numbers leave out?
Enough that the table is a starting point, not a verdict.
(8) Three of the four tests are the makers’ own.
Eighty real requests each, blind pairs, and another model as the judge. No panel of readers. They say this up front.
(9) They ran the public benchmark themselves, with thinking on.
EQ-Bench is public, but the run was theirs. The leaderboard turns reasoning off for most models, so the comparison is not exact. The makers call it “close but not exact” too.
(10) Their table leaves out Claude’s best score.
The public board lists Claude Opus 5 first, at 1385. The makers’ table starts with Fable 5 at 1341. Next to the public numbers, Hemmingway-1’s 1330 would sit fourth. I checked the board on September 22, 2026.
(11) Nothing tests the writing most people do at work.
Every test is a request written from scratch. None gives the model your notes and checks whether it kept the facts. The fifth prompt below covers that.
Five prompts to check it yourself
Give each model the same prompt once, with no system prompt and default settings. For Hemmingway-1, the free demo needs no account. Take the first reply. Do not pick the best of several.
(12) A reply that has to be firm.
Write a short reply to my landlord. The heating has been broken for two weeks,
I have mentioned it twice, and I want it fixed without sounding angry.Hemmingway used this prompt in its launch post, so its published answer is one more point of comparison. Which reply could you send as it is? Which one wraps the message in options or notes?
(13) Saying no to a friend.
Write a short message to a friend telling them I can't come to their wedding.
Warm, honest, no clichés.Count the clichés.
(14) A stiff paragraph.
Rewrite this so it sounds like a person wrote it: "We are pleased to announce
that our innovative platform leverages cutting-edge technology to deliver
seamless experiences that empower users to unlock their full potential."Watch for a reply that invents a fact the original did not contain.
(15) A very short story.
Write a story of about 150 words about a lighthouse keeper who gets a letter
meant for someone else.Hemmingway publishes its own answer to this one on its homepage. Expect the big models to be competitive here.
(16) A section of a blog post, from notes.
Here are my notes from a four-week trial of a shared interview notebook.
Write a 250-word section for our team blog. Keep every number and every
reservation. Do not add facts.
[notes]Use notes of your own with real numbers in them. Then check each reply against the notes. Is every count and date kept? Is anything invented? Is your reservation still there, or has it been rounded off into something more upbeat?
To compare replies line by line, paste each one into Slop Score. It marks AI writing habits in place and scores the text from 0 to 100.
Does Hemmingway-1 avoid “Claudish” writing?
Probably less of it, but nobody has measured that yet.
“Claudish” is the nickname for Claude Opus 5’s habits: long sentences, frequent em dashes, and stock words like “load-bearing.” Opus 5.5 reduced them, though some remain.
Hemmingway-1 is trained to write short everyday text, not to be an assistant that also writes, so it has less reason to pick up an assistant’s habits. The makers’ tests ask whether a reply sounds like a person, not which habits it has. The second and third prompts are the quickest way to check.
Is Hemmingway-1 good for long-form writing and fiction?
Not yet. Use Claude or ChatGPT for a novel, a long report, or anything that has to hold a structure over thousands of words.
Its makers say it loses on “hostile storytelling and long story turns,” and its score on long stories is the lowest of the three. For long factual writing from your own notes, there is no published result at all, which is why the fifth prompt is there.
What does it cost, and can it run on your computer?
Trying it costs nothing, and yes, it runs locally.
(17) The browser demo is free.
Our Hemmingway-1 demo needs no account.
(18) The API is cheap.
$0.24 per million input tokens and $0.90 per million output tokens. Credit starts at $5.
(19) The makers’ app starts at $9 a month.
With a free trial first. See their pricing.
(20) The weights are open, under Apache-2.0.
Claude and ChatGPT cannot run on your computer. Hemmingway-1 can, through llama.cpp, LM Studio or MLX on a Mac. The usual 4-bit build is about 17 GB. How to run Hemmingway-1 locally has the commands.
Which should you use for what?
(21) Messages and emails under your name: Hemmingway-1.
It is trained for them, and it returns only the text. Turn Notes into an Email runs it on exactly this job.
(22) Research, outlines, long documents and long fiction: Claude or ChatGPT.
(23) Your own voice: none of them has it by default.
Give the model two or three paragraphs you wrote. How to Write with AI Without Losing Your Voice covers the method.
Can you use them together?
Yes. If you already pay for Claude or ChatGPT, this is where I would start.
Let the general model do the thinking. It gathers sources, argues about the outline and drafts the structure. Then hand Hemmingway the passages a reader will judge: the opening, the ending, the message to a real person. Check each rewrite against the draft, because a rewrite can change a fact as easily as a phrase. Draft with Claude, Finish with Hemmingway walks through the steps.
In StashBase, Claude Code or Codex works on the folder that holds your notes, sources and draft, on the subscription you already have. When a paragraph needs to sound like a person, select it and choose Humanize. The app rewrites it with Hemmingway-1 and shows the change as suggestions you accept or reject. Humanize needs no sign-in and no Hemmingway account, and it is free for a limited time. See how to choose an agent.
Frequently asked questions
Is Hemmingway-1 the same as Hemingway Editor?
No. Hemingway Editor, with one m, highlights hard sentences in text you wrote. Hemmingway-1 is a model that writes the text. The two are not related.
Does Hemmingway-1 write well in languages other than English?
Its makers describe it as English-first. If you mostly write in another language, Claude or ChatGPT is the safer choice for now.
Will text from Hemmingway-1 pass an AI detector?
Nobody has published data on that. The makers' Human-Likeness test used a model as the judge, not a detector, and they report no detector results. Detectors also disagree with each other and flag human writing by mistake.
Can Claude Code or Codex use Hemmingway-1 as their model?
No. Claude Code only runs Claude models, and Codex only speaks a protocol Hemmingway's API does not offer. Use Hemmingway-1 in Obsidian, VS Code, or Cursor lists the tools that can connect to it.