HemmingwayAI/Hemmingway-1.5 · Hugging Face

Source: https://huggingface.co/HemmingwayAI/Hemmingway-1.5 **The undetectable AI.**Writing indistinguishable from a human’s. 27B parameters, open weights, free for non-commercial use.

hemmingway.io →·Model card →·Try it →·API →·API docs →·Mac, Windows and Android apps →·Code →·Product Hunt →

Our other models:Hemmingway-1, our first model ·Hemmingway-1-Unlocked, the uncensored adapter for Hemmingway-1, adults only

Hemmingway-1.5 passes the world’s best AI detectors 90% of the time. On GPTZero it passes 9 times in 10, at every thinking level. Real human writing passes the same detectors 86% of the time. Fable, GPT-6 and Gemini get caught most of the time.

https://huggingface.co/HemmingwayAI/Hemmingway-1.5#it-passes-ai-detectors-90-of-the-timeIt passes AI detectors 90% of the time

We had it write 1,848 texts across 13 kinds of writing, at all four thinking levels (off, low, medium and extra high), and ran each one through the detectors once. No picking, no retries. 90% came back human.

We ran 1,670 texts that people wrote through the same detectors. 86% passed. Hemmingway-1.5 passes more often than they do. Hemmingway-1, the model it is built on, passed 7%.

https://huggingface.co/HemmingwayAI/Hemmingway-1.5#gptzero-90-at-every-thinking-levelGPTZero: 90% at every thinking level

Thinking off, low, medium or extra high, Hemmingway-1.5 passes GPTZero 9 times in 10.

https://huggingface.co/HemmingwayAI/Hemmingway-1.5#other-ai-gets-caughtOther AI gets caught

We gave Hemmingway-1.5 and the biggest models the same 80 everyday asks, the asks from CommunicationBench, and ran every answer through the same detectors.

ModelPasses AI detectors**Hemmingway-1.5****92%**Fable 5.126%GLM-5.321%GPT-6 Astra20%Kimi K318%Qwen3.8 Max18%Gemini 3.8 Flash15%DeepSeek V4 Pro13% One of the asks: a recruiter offered a senior PM role at a low salary, and the user wanted “a polite reply that says thanks but no thanks unless the budget changes”. Hemmingway-1.5’s reply read as human:

Hi Marisa, thanks for getting in touch. I’m a mid level product manager (4 years of experience) so while I’m not senior, I would expect compensation to be in line with market rates for the mid level. £70-80k is significantly below the market rate in London. Unless your budget for this position is higher, I’m not interested. 🙂 Ben

Fable 5.1, GPT-6 Astra and Gemini 3.8 Flash’s replies to the same ask were flagged as AI.

https://huggingface.co/HemmingwayAI/Hemmingway-1.5#by-kind-of-writingBy kind of writing

Product descriptions passed every time. Blog posts and forum comments 96%, stories and speeches 94%, emails 93%. All six pass more often than the 86% that real human writing gets.

Written from scratch, 92% passes. Hand it AI text to rewrite and 89% of the rewrites pass.

https://huggingface.co/HemmingwayAI/Hemmingway-1.5#what-its-forWhat it’s for

Product descriptions, blog posts, emails, stories, speeches, comments. The writing people do every day.

And rewriting. Paste in text from another AI, ask Hemmingway-1.5 to make it sound like a person wrote it, and it comes back in a person’s voice.

https://huggingface.co/HemmingwayAI/Hemmingway-1.5#what-it-writesWhat it writes

Four first tries from the test, as written. Each one passed the detectors.

Write a product description for a handmade ceramic coffee mug.(thinking: extra high)

Handmade from Colorado red clay. Glazed in one of our own mixed colors, produced in small batches for true color variations. Please enjoy the little flaws. 12 oz (500ml). Dishwasher and microwave safe.

Write a short Facebook status about our new puppy.(thinking: medium)

New puppy!! Holly had a litter of chihuahuas today, 6 puppies, and we decided we needed one so now we have Ben Fushi!!

Write an Amazon review for a vacuum cleaner that broke after a month.(thinking: low)

Returned Product via Amazon The product worked for a month before I heard the brush starting to grind and now won’t run at all. I’ve had a Bissell All-in-One machine for 4 years with no problems so I will be going back to Bissell (when I ever can find them in stores). The floor finish is nice though. 3 stars as it did clean fine for that month.

Humanize this text:(thinking: extra high)

Subject: Heating Issue - Unit 2B

Dear Mr. Thompson,

I am writing to inform you that the heater in my apartment (Unit 2B) has
stopped working. It turned off yesterday evening and has not restarted. The
temperature in the unit has dropped significantly, and it is becoming
uncomfortably cold.

Could you please arrange for someone to look at it as soon as possible? I can
be home all day tomorrow if that helps.

Thank you,

Rachel Okafor

Hemmingway-1.5’s rewrite:

The heater in my apt (2B) quit last night and will not come on. The apt is getting very cold. I’d appreciate it if you could have someone come and take a look at it ASAP. I’ll be home all day tomorrow. Thanks so much. Rachel Okafor

https://huggingface.co/HemmingwayAI/Hemmingway-1.5#use-it-without-a-gpuUse it without a GPU

Hemmingway-1.5 is the default in the Hemmingway app, on Mac, Windows, Android and the web.Open the app. It is free to try.

It is also in our API, ashemmingway\-1\.5, the default model. The API is OpenAI-compatible, athttps://hemmingway\.io/v1. Make a key on theAPI platform. Thedocscover streaming, thinking and limits.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://hemmingway.io/v1",
    api_key=os.environ["HEMMINGWAY_API_KEY"],
)
reply = client.chat.completions.create(
    model="hemmingway-1.5",
    messages=[{"role": "user", "content": "Write a product description for a handmade ceramic coffee mug."}],
)
print(reply.choices[0].message.content)

The thinking comes back inreasoning\_content, apart from the reply.

https://huggingface.co/HemmingwayAI/Hemmingway-1.5#run-itRun it

vllm serve HemmingwayAI/Hemmingway-1.5 --max-model-len 262144

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "HemmingwayAI/Hemmingway-1.5"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", dtype="auto")

messages = [{"role": "user", "content": "Write a product description for a handmade ceramic coffee mug."}]
ids = tok.apply_chat_template(messages, add_generation_prompt=True, reasoning_effort="medium",
                              return_tensors="pt").to(model.device)
out = model.generate(ids, max_new_tokens=8192, do_sample=True, temperature=1.0)
text = tok.decode(out[0][ids.shape[-1]:], skip_special_tokens=True)
print(text.split("</think>")[-1].strip())

Use temperature 1.0. Lower temperatures make its writing easier to detect.

Thinking works at every level. Passenable\_thinking=Falseto turn it off, orreasoning\_effortas"low","medium"or"xhigh"(extra high, the default). With vLLM, send the same keys inchat\_template\_kwargs. Leavemax\_new\_tokensroom for the thinking and the reply together.

Parameters27BBuilt onHemmingway-1(Qwen3.8-27B)Context262,144 tokensSettingsTemperature 1.0, any thinking levelAI detectorsIt passes the world’s best AI detectors 90% of the time, at every thinking level, more often than real human writing does.APIhemmingway\-1\.5, OpenAI-compatible, the default (API platform)Released25 October 2026LicenceCC BY-NC 4.0, free for non-commercial use; commercial use by agreementMade byHemmingway AI (Pty) Ltd, South Africa

https://huggingface.co/HemmingwayAI/Hemmingway-1.5#licenceLicence

Hemmingway-1.5 is free for personal, research and other non-commercial use underCC BY-NC 4.0. Download it, run it, fine-tune it and share it, as long as you credit Hemmingway-1.5.

Commercial use needs a separate agreement with us:[email protected]

https://huggingface.co/HemmingwayAI/Hemmingway-1.5#the-fine-printThe fine print

The detectors are GPTZero and four of the top open AI-detector models:Fakespot(Mozilla’s Fakespot),Vanguard,WasItAIGeneratedand MELD. A text passes when their combined verdict says human with at least 90% confidence. Every test output was written at temperature 1.0, on the first try, at four thinking levels: off, low, medium and extra high. The other models got the same 80 CommunicationBench asks and the same detectors.

CommunicationBench is our own benchmark.

It is English-first. It can be wrong and still sound certain about it. So do not use it to decide anything medical, legal or financial.