If you write with Claude and want to try something else, the right alternative depends on what you write. For long fiction, the closest model we have tested is ChatGPT. For emails and messages that have to sound like you, try Hemmingway, which is ours. If you work in Gmail and Google Docs all day, Gemini is already there. For stories from outside the big assistants, GLM-5.3 and Kimi K3 are worth a look.
One thing first. In our own tests Claude is a very good writer, so "alternative" does not mean "better". Its Fable 5.1 model wrote the best long stories of any model we have tested. People usually look elsewhere for a narrower reason, and that reason tells you which tool to try.
The short answer
| Tool | What it is | Good for | Watch out for |
|---|---|---|---|
| ChatGPT | OpenAI's general assistant | Long stories, and short messages you can paste as they are | Scored below Claude on sounding like a person in our tests |
| Gemini | Google's general assistant | Writing inside Gmail and Google Docs | We have not tested it, so we have no scores |
| Hemmingway | Our writing app, on our own model | Emails, texts, replies and short pieces in your voice | Not for novels, code or research. English-first |
| GLM-5.3 | A model from Z.ai, which runs its own chat assistant | Stories | Wrapped its messages in notes in our tests |
| Kimi K3 | A model from Moonshot AI, in the Kimi assistant | Scenes that turn on how people feel | Also wrapped its messages. Its own site pitches it at coding and knowledge work |
Why people look past Claude
There are three usual reasons, and our tests bear out the first two.
- The wrapper. On eighty real email and message requests, Claude's Fable 5 put the message inside commentary, options or notes in more than nine replies out of ten. Fable 5.1 did it about two times in three, and its messages were the best quality on our board. You still have to cut them out before you can send them.
- It reads like a careful assistant. Asked which of two replies a person wrote, our judge picked Fable 5.1's less often than Hemmingway-1's: 1006 against 1032 on Human-Likeness, which is our own benchmark.
- You want a different kind of tool. Something that drafts replies from your inbox, something built into Google's apps, or a model you can download and run yourself.
If none of these is your problem, the honest advice is to stay where you are.
The alternatives, one at a time
ChatGPT
ChatGPT is OpenAI's general assistant. Its own site lists writing among its uses, from messages and notes to stories and posts, next to images, code and voice.
In our tests its model, GPT-6 Astra, was the second-best story writer on our board at 1312 on StoryBench, behind only Fable 5.1 at 1350. On everyday messages it scored 976 against Fable 5.1's 1024. But its messages were short, about 59 words on average, and it almost never wrapped them in commentary. If the wrapper is what bothers you about Claude, ChatGPT is the most direct swap. The cost is that its replies sounded less like a person: 964 on Human-Likeness. The two side by side.
Gemini
Gemini is Google's assistant. Google's site lists a free plan with a Google Account, and lists Gemini inside Gmail and Docs under its paid plans. In Gmail it can write an email from a prompt, tidy a rough draft and sum up a long thread.
We did not benchmark Gemini, so we cannot say how its writing compares with Claude's. The reason to pick it is where it lives: if your writing already happens in Google's apps, an assistant inside them saves the copying back and forth. Hemmingway vs Gemini has more.
Hemmingway
Hemmingway is our own app, built on our own 27B model, Hemmingway-1, which was trained for writing to people. In the same test of eighty requests it scored 1026 on everyday messages, level with Fable 5.1, and came first on Human-Likeness. It gives you the message and nothing around it.
The app works differently from a chat assistant too. You tell it who people are, and it writes to your boss, a client or your sister the way you would. In the apps for Mac, Windows and Android it reads your mail and chats and drafts each reply for you to send. The model has open weights under the CC BY-NC 4.0 licence, so you can run it yourself for non-commercial use.
GLM-5.3 and Kimi K3
These are the two other story writers we have tested. GLM-5.3 comes from Z.ai, which runs a chat assistant on its GLM models. It scored 1254 on StoryBench, ahead of Hemmingway-1 and behind Fable 5.1, GPT-6 Astra and Fable 5 Max.
Kimi K3 comes from Moonshot AI. It scored 1197 on StoryBench, level with Hemmingway-1, and 1332 on the public EQ-Bench 4, second only to Claude's Fable 5. That test measures how well a model reads people, which matters for fiction that turns on feelings.
On everyday messages both came in behind Fable 5.1 and Hemmingway-1, and both wrapped the message in more than nine replies out of ten. So neither fixes the wrapper.
Where Hemmingway does not fit
We make Hemmingway, so here is where Claude beats it.
- Novels and long chapters. Fable 5 Max scored 80 points above Hemmingway-1 on StoryBench, and Fable 5.1 is stronger still.
- Reading people. On EQ-Bench 4, Fable 5 scored 1341 and Hemmingway-1 1330.
- Everything that is not writing. Claude is a general assistant for code, analysis and long documents. Hemmingway is not.
Hemmingway is also English-first, and it can be wrong and still sound certain. Do not use it for medical, legal or money decisions.
How to choose
- Name what you want to change. If it is the notes around each message, try ChatGPT or Hemmingway. If it is the sound of the writing, try Hemmingway.
- If you write fiction, test ChatGPT on one chapter before you move. In our tests Claude still wrote the better long stories.
- If most of your writing happens in Gmail or Docs, try Gemini there, on real work, for a week.
- Give each tool the same three requests, with the same facts, and count how much you cut or change before you would send the result.
Every score on this page is on the model card or in our test results by job. CommunicationBench, Human-Likeness and StoryBench are our own benchmarks. EQ-Bench 4 is public.
Try HemmingwayDownload the app
Common questions
What is the best alternative to Claude for writing?
It depends on the writing. For long fiction, ChatGPT's GPT-6 Astra was the closest to Claude in our tests. For emails and messages, Hemmingway-1 was level with Claude's Fable 5.1 and gives you the message with no notes around it. If you write inside Google's apps, try Gemini, which we have not tested.
Is there a free alternative to Claude?
ChatGPT's site says its free version is open to everyone, and Google's site lists a free Gemini plan with a Google Account. Hemmingway has a free trial with no card needed; see /pricing/. Claude's own site lists a free plan too, so cost alone may not be a reason to switch.
Is ChatGPT better than Claude for writing?
Not in our tests. Claude's models came out ahead on everyday messages, sounding like a person, long stories and the public EQ-Bench 4. ChatGPT's GPT-6 Astra was the more direct of the two: short messages, with almost no commentary around them.
Which Claude alternative sounds most like a person?
In our Human-Likeness test, where a judge picks which of two replies a person wrote, Hemmingway-1 came first at 1032, ahead of Fable 5.1 at 1006 and GPT-6 Astra at 964. Human-Likeness is our own benchmark. We did not test Gemini.
Is there a Claude alternative I can run on my own computer?
Hemmingway-1 has open weights under CC BY-NC 4.0, free for non-commercial use, so you can download it and run it on your own hardware. It is a 27B model built for writing, not a general assistant. How to run it locally.