Most AI writing tests measure essays, code or stories. Most of what people actually write is shorter: an email to a client, a text to a landlord, a note to a colleague, the message you have been putting off. We built a test for that.
How the test works
CommunicationBench is eighty real requests of the kind people bring to an assistant. Every model answers all eighty. Then every answer goes head to head with another model's answer to the same request, and a judge picks the better one. The pairs are shuffled and run in both orders, so position cannot sway the result, and the judge is a different model from the ones being judged. Scores are on an Elo scale.
It is our own benchmark. We built it and ran it, and we say that up front on Hemmingway-1's model card. The table below is Hemmingway-1's, our first model.
The results
| Model | CommunicationBench |
|---|---|
| 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 |
Grok 4.6 and DeepSeek V4 Pro also came in behind Hemmingway-1. The top two are close: Hemmingway-1 and Fable 5.1 are level within the margin. Hemmingway-1 is fifty points ahead of GPT-6 Astra.
The same eighty asks, through AI detectors
Our newest model, Hemmingway-1.5, released on 25 October 2026, is built on Hemmingway-1 and is now the default in the Hemmingway app. We gave it and the biggest models the same eighty CommunicationBench asks, and ran every answer, first try only, through GPTZero & Pangram and four of the top open AI detectors.
| Model | Passes AI detectors |
|---|---|
| Hemmingway-1.5 | 92% |
| Fable 5.1 | 26% |
| GLM-5.3 | 21% |
| GPT-6 Astra | 20% |
| Kimi K3 | 18% |
| Qwen3.8 Max | 18% |
| DeepSeek V4 Pro | 13% |
In the full detector test, 1,848 texts in 13 kinds of writing, Hemmingway-1.5's emails passed 93% of the time. Real human writing passes the same detectors 86% of the time. How the test was run.
Where the differences are
Averages hide the interesting part.
- Hard asks. The messages you keep rewriting: saying no, asking for money back, telling someone something they will not like. Here the gap is widest. Judged on which reply sounds like a person wrote it, GPT-6 Astra's won 9% of the time and Hemmingway-1's 72%.
- Money and admin, work, talking someone round. Hemmingway-1 wins these, most by a wide margin.
- The wrapper. Fable 5, GLM-5.3 and Kimi K3 bury the message inside commentary, options or notes in more than nine replies out of ten. You then have to dig the email out before you can send it. Hemmingway-1 gives you just the message.
- Where Hemmingway-1 loses: hostile storytelling and long story turns. The story models are better there.
Which to use
For everyday email and messages, Hemmingway-1 and Fable 5.1 lead our CommunicationBench, and Hemmingway is the one that reads as a person wrote it: Hemmingway-1 came first in our Human-Likeness test, and 92% of Hemmingway-1.5's answers passed AI detectors, against 26% of Fable 5.1's. Both are 27B models with open weights.
The Hemmingway app writes with Hemmingway-1.5, and it can also read your mail and chats and draft each reply for you. How that works.
Try HemmingwayDownload the app
Common questions
What is the best AI for writing emails?
In our CommunicationBench, eighty real email and message requests judged blind, Hemmingway-1 scored 1026 and Fable 5.1 1024, level within the margin. Fable 5, GLM-5.3, Kimi K3 and GPT-6 Astra followed. On the same asks, 92% of Hemmingway-1.5's answers passed AI detectors, the most of any model we tested. Both tests are ours, and we say so on the model cards.
Is ChatGPT good at writing emails?
GPT-6 Astra scored 976 in our test, fifty points behind Hemmingway-1, and AI detectors passed 20% of its answers against 92% of Hemmingway-1.5's. Its messages were short and it rarely wrapped them in commentary. But on hard asks, judged on which reply sounded like a person wrote it, its replies won 9% of the time against Hemmingway-1's 72%.
Why does AI add notes and options around the email?
Many models wrap the message in commentary, several options or a note on the tone. In our test Fable 5, GLM-5.3 and Kimi K3 did it in more than nine replies out of ten, so you have to dig the email out before sending. Hemmingway-1 gives you just the message.
Can AI answer my emails and messages for me?
In the Hemmingway apps it can read your mail and chats and draft each reply for you to send. How that works.