Hemmingway Try Hemmingway

Blog · Comparison

Character.AI alternatives

Character.AI alternatives fall into three kinds: other character apps, general assistants and open models. What each is good for, and where Hemmingway fits.

By the Hemmingway team ·

Character.AI alternatives come in three kinds. There are other character and companion apps, which work much like it: pick or make a character and chat. There are general assistants, such as ChatGPT, Claude and Gemini, which will play a character if you set one up in the conversation. And there are open models, which you run on your own computer or reach through an API, usually with a separate chat app on top. Which one suits you depends on whether you want a ready-made app, the strongest writing, or control over where your chats go.

We make Hemmingway, and it is not a character app. Its model, Hemmingway-1, was built for talking to people and for writing. It scores well on reading people, with 1330 on the public EQ-Bench 4, but in our own tests it is weaker at hostile storytelling and long story turns, where story models do better. This page explains the options and says where Hemmingway fits, and where it does not.

What Character.AI is, and why people look elsewhere

Character.AI is a service for chatting with AI characters, many of them made by other users. You pick a character, or write one, and talk to it. People look elsewhere for better writing, more control over the model, or chats kept on their own machine.

The three kinds of alternative

KindExamplesGood forTrade-off
Character and companion appsOther character-chat apps, companion apps such as ReplikaReady-made characters, no set-upYou use the app's model and its rules
General assistantsChatGPT, Claude, GeminiStrong writing, one tool for many jobsYou set the character up yourself; they are not built around characters
Open modelsHemmingway-1, and others with published weightsControl, privacy, running it yourselfNeeds hardware, or an API key and a chat app

We have not tested Character.AI, other character apps or companion apps, so we cannot say which of them writes best. What we can offer is test results for the models themselves, which is the part underneath every one of these.

Other character and companion apps

These are the closest to Character.AI. You get an app, a list of characters and a chat window, with little or nothing to set up.

Suits: people who want the Character.AI experience and are happy for the app to choose the model.

Worth checking before you move: who can read your chats, whether they are used for training, and whether you can export them. Each app answers those differently, so read its privacy policy.

General assistants

ChatGPT, Claude and Gemini are general assistants. None of them is a character platform, but each will play a character if you describe one in the conversation. Their strength is the model: the largest ones are very capable writers. In our StoryBench, our own creative-writing test, the top scores went to Fable 5 Max, from Claude's family, and GLM-5.3.

Suits: people who want strong storytelling and do not mind setting up a character themselves each time.

The catch: they are built to be helpful assistants, and that can show. A character can drift back into an assistant's voice, or the model can add notes and options around what the character says. Clear instructions at the start help. More on how the big assistants compare for writing. We did not test Gemini.

Open models

Open models publish their weights, so anyone can run them. You can run one on your own hardware, where your chats never leave the machine, or reach one through an API. Either way you usually pair it with a separate chat app that talks to the model.

Suits: people who want control over the model and its settings, people who care most about privacy, and people who like to tinker.

The catch: running a large model yourself needs serious hardware, and setting up a chat app takes some effort. How to run an open model for roleplay.

Where Hemmingway fits

Hemmingway is two things, and they fit in different places on this list.

Hemmingway-1, the model, is an open model. It has 27 billion parameters, a 262,144-token context, and open weights under Apache-2.0. You can run it yourself, or use it through an OpenAI-compatible API as hemmingway-27b. That puts it in the third row of the table.

The Hemmingway app is not a character app. It runs on Mac, Windows, Android and the web, and it is built for writing to real people: messages, emails, letters and short stories. You can ask it to play a character in a chat, as you can with any assistant, but it has no character library and is not designed around characters.

Here is how the model scores on the two things that matter most for character chat.

TestWhat it measuresHemmingway-1Top of the table
EQ-Bench 4 (public)Emotional intelligence in conversation1330, thirdFable 5, 1341
StoryBench (ours)Creative writing1197, level with Kimi K3Fable 5 Max, 1277

EQ-Bench 4 is a public benchmark; we ran it ourselves with its official harness. StoryBench is our own benchmark: blind, run in both orders, with a judge that is not one of the models being judged. All the scores are on the model card.

What that means in practice:

  • Where it is good: characters who have to read a mood and answer a person well. Hemmingway-1 is third on EQ-Bench 4, inside twelve points of the top, and it is a 27B model among much larger ones.
  • Where it is weaker: hostile storytelling and long story turns. In our own breakdown by request type it loses on both, and the story models are better at them. If your roleplay is long adventures with a lot of narration in each reply, a large story model will serve you better.
  • What it is not: a character platform with ready-made characters.

Try HemmingwayDownload the app

How to choose

  1. You want the same kind of app with different characters. Try another character or companion app. Check its privacy policy first.
  2. You want the best storytelling and will set things up yourself. Try a large general assistant, or a large story model. Fable 5 Max and GLM-5.3 scored highest on our StoryBench.
  3. You want your chats on your own machine. Run an open model yourself. Hemmingway-1's weights are published under Apache-2.0.
  4. You want character conversations that feel emotionally right, in an open model. Hemmingway-1 is worth trying: it is open, and third on EQ-Bench 4. Expect it to be weaker on long, dark story turns.

Moving a character to a new model

Whatever you switch to, you can bring a character with you by writing it down. Put these in the first message, or in the system prompt if your app has one:

  • the character's name, age and situation;
  • how they speak, with one or two sample lines;
  • what they want and what they would never do;
  • who you are playing, and that the model must not write your character's words or actions;
  • how long each reply should be.

This works across apps, assistants and open models, because it is plain text. More on getting good roleplay from any model.

Common questions

What is the best alternative to Character.AI?

It depends on what you want. For the same kind of experience, another character or companion app; for the strongest writing, a large general assistant or story model; for privacy and control, an open model you run yourself. We have not tested Character.AI or other character apps, so we cannot rank them.

Is Hemmingway a Character.AI alternative?

Partly. Hemmingway-1 is an open model that scores 1330 on the public EQ-Bench 4, so it is a fair choice if you run models yourself or through an API. The Hemmingway app is a writing app, not a character platform, and in our tests the model is weaker at hostile storytelling and long story turns than story models.

Can ChatGPT or Claude do roleplay?

Both will play a character if you describe one in the conversation. They are general assistants rather than character apps, so you set the character up yourself, and they can drift back into an assistant's voice. Clear instructions at the start help.

Is there an open-source alternative to Character.AI?

Open models, Hemmingway-1 among them, have published weights you can run yourself, usually with a separate chat app on top. Hemmingway-1 is released under Apache-2.0. How to run one.

Read next

  • The best AI for roleplayThe best AI for roleplay stays in character, reads emotion and lets you steer. What to look for, the scores that matter, and where our own model falls short.
  • Open-source models for roleplayWhy people run open-source models for roleplay, how to set one up with vLLM or an OpenAI-compatible API, and how Hemmingway-1 compares for character chat.
  • The best ChatGPT alternatives for writingThe best ChatGPT alternatives for writing, ranked by use case: Claude for long fiction, Hemmingway for messages, GLM-5.3 and Kimi K3 for stories.
  • EQ-Bench 4: how Hemmingway-1 scoresEQ-Bench 4 tests emotional intelligence in conversation. We ran it on Hemmingway-1 with the official harness: 1330, third of 27, ahead of GPT-5.5 and Opus.