Building nownothing released yet

The open model for marketing work

Doctors have an AI model built for medicine. Lawyers have one for law. Coders have several. Marketers have none. We are building one in the open and giving it away.

Free to download. Free to use. Free to sell what you build with it.

The gap

Every serious field has its own model. Marketing does not.

More than half of company AI models will be built for one field by 2027, according to Gartner, up from 1% in 2024. Marketing is the biggest discretionary spend in business, and the tools marketers use, Jasper and Copy.ai among them, all rent a general model from someone else.

MedicineMed-PaLM, BioGPT, BioMedLM
LawChatLaw, DISC-LawLLM, PKUlaw
FinanceBloombergGPT, FinGPT
CodeCode Llama, DeepSeek Coder
ScienceGalactica
Marketingnothing open exists

What it does

Five jobs marketers do every day

It will not out-think Claude or Gemini, and we will not pretend it does. It is built to do a handful of repetitive jobs quickly, cheaply, and the same way every time.

creative_tag

Tag ad creative

Show it an ad. It tells you the hook, the angle, who it speaks to, what it offers, and where it sits in the funnel. Always the same labels, so you can count them.

ad_copy

Write ad copy

Say what you are selling and to whom. Get a headline, body text, and button label that already fit the platform's character limits.

metric_diagnosis

Diagnose a campaign

Paste your numbers. It names what is broken and what to fix first. If there is too little data to tell, it says so rather than guessing.

lead_qualify

Score a lead

Give it what you know about a lead. Get a score, a band, and the exact signals behind it. It never invents a reason it cannot point to.

chat

Answer questions

Ask anything about marketing. Get a direct answer, or a question back when a number it needs is missing.

ad_policy

Predict ad rejectionsmaybe

Would this creative be rejected, and why. Nothing does this today. We are still deciding whether it makes the first release.

Not built yetthis code does not run yet

Works with the tools you already use

It speaks the same language as the OpenAI library, so switching to it means changing one line. Or skip us entirely, download the model, and run it on your own computer.

python
from openai import OpenAI

client = OpenAI(base_url="https://api.gloofy.ai/v1")

client.chat.completions.create(
    model="gloofy-1-nano",
    messages=[{"role": "user",
               "content": "Tag: Still paying 3% on card payments?"}],
)

# {"hook": "question", "angle": "pain_relief",
#  "persona": "business_owner", "offer": "lead_magnet",
#  "funnel_stage": "consideration"}

The numbers

Every specification, and what each one means

Model makers publish a table like this and rarely explain it. Here is ours, with plain English underneath. None of these are measured yet, so treat them as the target we are building towards.

Specificationgloofy-1gloofy-1-nanogloofy-0
Parameters8.2B1.7B561M
ArchitectureDenseDenseDense
Context window32,768 tokens32,768 tokens2,048 tokens
Vocabulary151,936151,93665,536
Download (4-bit)~5 GB~1.1 GB~0.6 GB
Runs on a laptop16 GB RAM8 GB RAMAnything
LicenceApache 2.0Apache 2.0Apache 2.0
Built fromQwen3-8BQwen3-1.7BNothing

Parameters

The adjustable numbers inside the model, learned during training. More usually means smarter, but slower and dearer to run. Ours has 8.2 billion. The largest open model today has 2.8 trillion.

Context window

How much text it can hold in mind at once. 32,768 tokens is roughly 90 pages. Go past it and the earliest text falls out of view.

Tokens

Models cannot read letters. Text is chopped into word-sized pieces first, and a token is one piece. Roughly four characters of English each.

Dense vs mixture-of-experts

Dense means every parameter is used for every word. Mixture-of-experts models switch on only a slice at a time, which is how a 2.8-trillion model can be affordable to run. Ours is dense because at this size it is simpler and faster.

Quantization

Storing each number with less detail so the file shrinks. Four-bit makes the download about four times smaller and costs a little accuracy. It is why an 8-billion model fits on a laptop.

Apache 2.0

A licence that lets you use it commercially, change it, and ship it inside your own product, without asking us or paying anyone.

Models

Three sizes, one of them a toy

Named now so the plan can be judged later. The weights go on Hugging Face for anyone to download, because a model you cannot have is not infrastructure.

gloofy-1

8.2B

The main one. Does all five jobs and holds a conversation.

Not built yet

gloofy-1-nano

1.7B

Sorting and labelling only. Small enough to run on a laptop, fast enough to tag a thousand ads while you wait.

Not built yet

gloofy-0

561M

Built from nothing as a way to learn how. Published because the process is worth showing, not because you should use it.

Building now

marketing-bench

The test comes out before the model

Medicine has a shared exam for AI models, written by 262 doctors. Law has one. Marketing has nothing, so we are writing it, and publishing it first so we cannot quietly change the questions to suit our own answers.

Published before the model

The first scores recorded are from models we did not train. Any improvement later is measured against a number that existed before we tried.

Shows where we lose

Claude and Gemini sit on the same leaderboard. Where they beat gloofy, the table says so.

Wider than our model

It covers marketing work in general, not only the parts we happen to be good at. A test written around its own author is advertising, not measurement.

Anyone can rerun it

The questions and the scoring code are public. If you doubt a number, you can check it yourself.

Build log

Built in the open, from the first line of code

This began by writing a language model from scratch on a laptop, because you should not publish one without knowing what is inside it. Every step is written up with its real numbers, including the parts that failed.

loss 3.66 to 1.57

A tokenizer, 50 million tokens, and a model that tells stories

Wrote the text-splitting algorithm from scratch, then trained six model designs against each other and let the results pick the winner. Wide and shallow beat deep and narrow. The winner trained for six hours and started writing stories with a beginning, a middle, and an end.

3.2M parameters, 287 seconds

A language model from absolutely nothing

A small model trained on Shakespeare, on a laptop, in under five minutes. Nobody told it that plays put speaker names in capitals. It worked that out on its own from one megabyte of text.

Promises

Written down early so we can be held to them

The weights stay free

Published on Hugging Face for anyone to download. If we ever switch the website off, the model still works on your own machine.

No paywall

There is no paid plan and none is planned. If that ever changes, the free downloads do not.

No customer data

Nothing from any company or customer is used to train it. Every example is written by us or already public.

Honest about its size

This is a small model. It will not beat Claude or Gemini at general thinking, and this site will never suggest it does.