Hemmingway-1

Hemmingway-1

Writes everyday messages and emails as one ready-to-send text instead of a list of options. A 27B model built on Qwen3.8-27B, free for non-commercial use, for people who want writing that sounds human.

Made by HemmingwayAI on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Hemmingway-1 is a 27B text model built on Qwen3.8-27B for the writing people do every day, such as messages, emails and awkward notes to a colleague. Where many models answer with options and an explanation, it returns the text itself. The weights are open and free for non-commercial use, and commercial use needs a separate agreement.

Who it suits: It suits people who write a lot of short messages, and developers who build writing helpers on an open model.

Industry | Who uses it | How they use it | Example
Personal and office writing | Anyone with an awkward message to send | Ask for a text that is ready to send | Request a message to a landlord about a broken boiler
Software development | Developers who build writing helpers | Serve the model with vLLM | Start the server with a context of 262,144 tokens
Research | People who compare writing models | Read the authors' benchmark results | Check how it did on the public EQ-Bench 4

If you need commercial use without an agreement, or advice on medical, legal or money decisions, you can skip it.

Where to use it:

• A vLLM server on your own GPU machine
• Python scripts that use the transformers library
• The hosted try-it page, plus Mac and Android apps, which the card links to
• Windows, macOS or Linux with Python, to download the files

Use case | Who | What they get back
Everyday messages | Anyone | One text to send, with no options or commentary around it
Work emails | Office staff | Drafts for awkward notes and requests
Hard asks | People who keep rewriting | A message that tries to talk the reader round

How to set it up: What you need first:

• Python and the Hugging Face command line
• GPU memory for about 53.8 GB of BF16 weights, or about 13.4 GB at 4-bit, which is an estimate
• vLLM or the transformers library

1. Install the Hugging Face tools with pip.
2. Download the model files to your computer.
3. Start it with vLLM or load it in Python.

pip install -U huggingface_hub
hf download HemmingwayAI/Hemmingway-1

How to use it: Start a vLLM server, or load the model in Python and send it a chat message. The card's own examples use the Altworld copy of the weights.

vllm serve Altworld/Hemmingway-1 --max-model-len 262144

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Altworld/Hemmingway-1"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", dtype="auto")

messages = [{"role": "user", "content": "Write the text I send my landlord about the broken boiler."}]
ids = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(ids, max_new_tokens=512)
print(tok.decode(out[0][ids.shape[-1]:], skip_special_tokens=True))

Tool | What it does
vllm serve | Starts a server for the model with a context of 262,144 tokens
transformers script | Loads the model, applies the chat template and prints the reply

Example: You ask for a message to your landlord about a broken boiler. The model answers with the message itself, ready to send, and not with three options and a paragraph about them.

For newcomers and experts: A good first try is to ask for one short email you have been putting off, then read how it sounds. If it needs fixing, say what to change and ask again.

For experts, this is a 27B model on the Qwen3.8-27B base with a 262,144-token context. The authors compared it in blind head-to-head matchups, run in both orders, with a different model as the judge. CommunicationBench, Human-Likeness and StoryBench are their own benchmarks, and EQ-Bench 4 is public. The trade is storytelling, where the card admits it falls behind on hostile storytelling plus long story turns.

You describe the message you need
The model reads it with its chat template
It writes the text without options or notes
You read it and send it

Good to know:

Title: Hard asks, picked as human-written
GPT-6 Astra: 9%
Hemmingway-1: 72%

• Benchmarks: the authors built three of the four tests themselves, and the card says so up front. On the public EQ-Bench 4 it placed third.
• Weak spots: hostile storytelling and long story turns.
• Reliability: it can be wrong and still sound sure, so do not use it for medical, legal or financial decisions.
• Language: English first.
• License: cc-by-nc-4.0, which allows non-commercial use only. Commercial use needs a separate agreement with the authors.
• Cost to run: no API key or paid service named for the open weights. The card does not state what the hosted page or apps cost.
• Data it reaches: it downloads files from Hugging Face. The card does not describe data handling in the hosted page or apps.
• Not for: commercial use without an agreement, and medical, legal or financial decisions.

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 26.9B
• Weights: about 53.8 GB in BF16; about 13.4 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 3,631
• Likes: 902
• License: cc-by-nc-4.0
• Languages: en
• Based on: Qwen/Qwen3.8-27B
• Quantized versions: 54
• Fine-tunes: 8
• Tags: transformers, safetensors, qwen3_5_text, text-generation, qwen3.8, chat, creative-writing, altworld, conversational

Links

• Model: https://huggingface.co/HemmingwayAI/Hemmingway-1
• Files: https://huggingface.co/HemmingwayAI/Hemmingway-1/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download HemmingwayAI/Hemmingway-1

Then follow the usage in its model card. Nothing in this kit runs by itself.