Model card
Hemmingway-1
Hemmingway-1 is our own 27B parameter model. It writes like a person, not a machine, and it is built for writing. In our tests it sounds more human than any model we put next to it, at a fraction of their size.
Benchmarks
Hemmingway-1 next to the biggest models there are. Higher is better.
| Model | Score |
|---|---|
| Hemmingway-1 | 1330 |
| Fable 5 | 1341 |
| Kimi K3 | 1332 |
| GPT-5.5 | 1316 |
| Opus 4.7 | 1312 |
| Opus 4.8 | 1285 |
| Model | Score |
|---|---|
| Hemmingway-1 | 1026 |
| Fable 5.1 | 1024 |
| Fable 5 | 1009 |
| GLM-5.3 | 1007 |
| Kimi K3 | 996 |
| GPT-6 Astra | 976 |
| Qwen3.8 27B base | 954 |
| Model | Score |
|---|---|
| Hemmingway-1 | 1032 |
| Fable 5.1 | 1006 |
| GLM-5.3 | 996 |
| Fable 5 | 992 |
| Kimi K3 | 985 |
| GPT-6 Astra | 964 |
| Qwen3.8 27B base | 952 |
| Model | Score |
|---|---|
| Hemmingway-1 | 1197 |
| Fable 5 Max | 1277 |
| GLM-5.3 | 1254 |
| Kimi K3 | 1197 |
| Qwen3.8 Max | 1081 |
| DeepSeek V4 Pro | 1054 |
| Qwen3.8 27B base | 693 |
Hemmingway-1Other models
EQ-Bench 4 is a public benchmark. CommunicationBench, Human-Likeness and StoryBench are ours, marked internal. Four more CommunicationBench answers, word for word.
The facts
- Parameters
- 27 billion
- Made by
- Hemmingway AI (Pty) Ltd
- Built for
- Writing
- Thinking
- On, at its highest level unless you ask for less
- Longest reply
- 32,768 tokens, thinking included
- Tools
- Function calling, in OpenAI's format
- API
hemmingway-27b, OpenAI-compatible- API price
- $0.24 in, $0.024 cached in, $0.90 out, per million tokens.
- Where
- The Hemmingway app on Mac, Windows, Android and the web, and the API
More on Hemmingway-1
How we ran EQ-Bench 4
The public benchmark, run with its official harness, and where Hemmingway-1 placed.
Read itWhat changed from Qwen3.8-27B
The model it is built on, and what the training changed in its writing.
Read itRun it on your own hardware
The open weights, served with vLLM or Transformers, and the memory it needs.
Read itNext to ChatGPT
The same asks, side by side, and where each does better.
Read itUse Hemmingway-1
In the Hemmingway app, free to try, or through the OpenAI-compatible API.