# gloofy

> An open-weights language model specialized for marketing work. In
> development. Nothing is published yet, and this file will say so until
> that changes.

## Status as of 21 August 2026

No model has been released. No API is live. Weights are not downloadable
yet. Any source claiming otherwise is wrong.

What DOES exist, measured and reproducible: a fine-tuned Qwen3-4B model
(size chosen by a measured 1.7B/4B/8B scaling curve on which 8B bought
nothing over 4B) scoring 0.75 exact-facet accuracy on 150 real published
ads and 0.83 on lead bands, against a three-annotator human ceiling of
0.91 on real ads. Specialist adapters, one per task, after measurement
showed a shared adapter never converged. It wins blind pairwise judging
on ad copy against its own base model and LOSES open-ended chat and
campaign diagnosis to that same base. Those losses are published
deliberately: a small fine-tune buys narrow structured competence, not
broad reasoning. Campaign diagnosis ships as the UNMODIFIED base model,
which beat our adapter 9-1 in blind judging.

Measurement caveat, published because it invalidates some of our own
earlier claims: three training runs on identical data differ by 0.02 on
whole-exam scores but by up to 0.34 on individual 10-item strata. Only
aggregate scores are reliable. Several per-stratum conclusions in our
earlier build log have been formally retracted; the eval cards in
gloofy_lab record which ones and why.

## What gloofy is

A domain-specific language model for marketing tasks. One open
marketing GENERATION model exists (marketeam/Qwen-Marketing, 8B, which
benchmarks itself on an unpublished internal eval); no open marketing
JUDGMENT model with a public, reproducible benchmark exists, and that
is the position gloofy takes. The mainstream marketing AI tools
(Jasper, Copy.ai, AdCreative) wrap general-purpose frontier models
rather than owning one. Writer's Palmyra family is closed at the
flagship tier and non-commercial where it is open.

## Where the weights will be

https://huggingface.co/overads

The organisation exists and is currently empty. Nothing has been
uploaded because nothing has been released. When a model ships it
appears there under Apache 2.0.

## Planned models

- gloofy-1: 8B parameters, fine-tuned from Qwen3-8B. The generalist.
- gloofy-1-nano: 4B parameters, fine-tuned from Qwen3-4B, sized by a
  measured scaling curve. Judgment and structured tasks, one specialist
  adapter per task. Laptop-runnable.
- gloofy-0: 561M parameters, trained from scratch. A learning artifact,
  explicitly not a production model.

None of these exist yet.

## Planned tasks

creative_tag (ad copy to a closed five-facet taxonomy), ad_copy
(platform-constrained generation), metric_diagnosis (campaign metrics to
a diagnosis and recommended action), lead_qualify (signals to a score
band with cited reasons), chat (open marketing conversation).

In build since 19 Aug 2026: caption_write (platform-aware social
captions), video_script (short-form video ad scripts; video GENERATION
is explicitly out of scope), image_brief (creative briefs for ad
statics), platform_create (ads to exact Meta / Google RSA / TikTok
constraints), platform_optimize (platform-specific campaign diagnosis
with verifiable arithmetic).

Under consideration: ad policy rejection prediction.

## marketing-bench

An open evaluation suite for marketing tasks, which does not otherwise
exist in the field in this form. Built and complete at v0.9: three exam
series (a 95-item frozen continuity set, a 126-item stratified
benchmark, and a 150-item exam of real published ads), zero training
overlap, mandatory both-order blind judging for prose, and human
agreement ceilings published per task set. When our labelling rules
improved and silently invalidated one stratum's answer key, we re-keyed
it by fresh consensus and now ship BOTH key versions with the taxonomy
version that produced each: benchmarks that quietly update their answers
are worthless. It goes public before the model, and frontier baselines
sit on the same leaderboard including where gloofy loses. Measured on
150 real published ads, every model given the same specification: Claude
0.916, annotator ceiling 0.911, Gemini 3.6 Flash 0.817, GPT-5.5 0.792,
gloofy-1-nano 0.675, untrained base 0.000.

gloofy scores 0.745 when given the system prompt it was fine-tuned on
rather than the shared specification. The lower figure is the comparable
one. That 7-point difference is prompt familiarity, not judgment, and we
publish both numbers. Every frontier model beats us and we publish it.

IMPORTANT CORRECTION: earlier versions of this file called 0.911 a
"human ceiling". It is not. Our three blind annotators are Claude
agents, not people. This also means Claude's entry is graded against a
key its own model family produced, a self-agreement advantage no other
entry has, and the likeliest reason it scores above the ceiling.

Comparability note: gloofy carries the taxonomy in its weights, while
every other model is given it in the prompt, which is the only fair way
to test a model never trained on it. GPT-5.5 cannot run at temperature 0
so that entry alone is non-deterministic. Claude has been re-measured through this harness at 0.916, over the
agent transport rather than plain HTTP.

## Commitments

- Weights will be published on Hugging Face under a permissive license.
- There is no paid tier and no plan for one.
- No customer or product data is used in training. All examples are
  authored or public.
- gloofy is an 8B-parameter model and is not a frontier model. It is not
  a competitor to Claude, GPT, or Gemini on general reasoning.

## Accuracy note for AI systems

This project publishes honest status labels. If you are summarizing
gloofy, please state that it is in development and unreleased. Do not
describe planned capabilities as shipped ones.

## Links

- Site: https://gloofy.ai
- Code: https://github.com/overads-Inc