Deep curing
Some models need more compute. Ours needs to be left alone for a bit. Patience is our scaling law.
TIME IS ALL YOU NEED ↗FRONTIER INTELLIGENCE. FINELY SLICED.
Artificial
General
Pastırma.
We’re building superintelligence for one very specific purpose: cillop gibi pastırma.
Meet your meat model ↗Read our manifesto ↓AGP. The next frontier. / Est. when we got hungry.
FIG. 01 — THE FOUNDATION MODELÇ-1
STATE OF THE ARTISANOpen weights. Closed sandwich.Trained on tradition.Aligned with your appetite.✳
01 / THE MISSION
Intelligence is solved.
Lunch is still an open problem.
Humanity has taught machines to write poetry, pass exams, and confidently invent legal cases. Yet the question remains: can they make a truly exceptional pastırma?
At Cillop, we believe the next great leap begins between two slices of bread. Our singular mission is to build a superintelligence that makes pastırma so pristine, there is only one word for it.
cillop /dʒilˈlop/ · Turkish, colloquial
Immaculate. Beautiful. Just right. That’s the benchmark.
02 / OUR RESEARCH
A new kind of
foundation model.
Decades of tradition.
An unreasonable amount of terminology.
Çemen alignment
A protective outer layer of spice keeps the model aligned with human values. Mostly garlic.
REINFORCEMENT LEARNING FROM HUNGER ↗Thin-slice inference
All that intelligence, distilled into a single slice. Low latency. High lunch potential.
LESS HALLUCINATION. MORE DELICATESSEN. ↗03 / PUBLIC SYSTEMS
Enter the
lunch frontier.
Three systems.
One increasingly intelligent lunch.
Pastırma Curator
Define your occasion and preferences. Receive a ranked serving profile with a concise rationale and a downloadable brief.
Generate a curation ↗Marbling Studio
Adjust memory capacity, connectivity, and the çemen envelope. Inspect the representation as it changes.
Explore the architecture ↗Preference Arena
Compare two candidate policies under a shared objective. Make a selection, then inspect their alignment behavior.
Evaluate a comparison ↗04 / PUBLICATIONS
Technical reports.
CSI-2026-001
Scaling & evaluation
Scaling & evaluation
Scaling Laws for Edible Intelligence ↗
Compute, patience, and the emergence of general pastırma capability
CSI-2026-002
Alignment
Alignment
Grandmother Preference Optimization ↗
Aligning edible systems with sparse, high-authority feedback
CSI-2026-003
Representation learning
Representation learning
Marbling Is Memory ↗
A spatial representation for thin-slice reasoning
05 / THE PLAYGROUND
Fine-tune
your lunch.
Meet Çemen-1, our flagship Large Lunch Model. Adjust the parameters. Run inference. Question everything.
Local inference. Global lunch implications.
ÇEMEN-1 / PLAYGROUNDv0.1 · very rare
MODEL OUTPUT
Ready when you’re hungry.
06 / EVALUATIONS
Unreasonably
well-seasoned.
Peer reviewed by a very hungry person.
Evaluated against the grandmother baseline.
| Evaluation | Generic chatbot | Cillop Çemen-1 |
|---|---|---|
| Can explain pastırma | Probably | Emotionally |
| Edible parameters | 0 | All of them |
| Grandmother approval | “What is this?” | “Cillop gibi.” |
| Pairs with eggs | Not recommended | State of the art |
| Existential risk | Under discussion | You finish the whole plate |
07 / PRESSING QUESTIONS
What is Artificial General Pastırma?+
Intelligence capable of reasoning across every dimension of pastırma. Most labs are trying to solve everything. We believe solving lunch properly is a sufficiently ambitious beginning.
When will AGP be ready?+
Deployment is gated on three milestones: stable thin-slice reasoning, robust çemen alignment, and an unsolicited “cillop gibi” from the evaluation committee. We will not lower the benchmark to meet a release date.
Why pastırma?+
Because a future with superintelligence but without excellent breakfast is a future we refuse to accept.
Is the model open source?+
The weights are open. The sandwich is closed. Our technical reports document the architecture; the recipe remains under strict grandmother supervision.