About
KittyHawk Model Hub is a small, volunteer-maintained catalog of open-weight large language models. We exist for one reason: it's surprisingly hard to get a clear, honest, single-page summary of any given model — its size, its license, what it's actually good at, and where to download the official weights.
Vendor pages are marketing surfaces. Aggregator sites are search-engine-optimized to within an inch of their lives. Leaderboards measure narrow capabilities and miss the practical questions: will this fit on my GPU? Can I use it commercially? Is the official tokenizer the same one the inference framework expects?
The catalog is small on purpose. Fifteen models is enough to cover the practical decision space; a thousand models is enough to make every page useless. We re-review each entry roughly twice a year, and retire models when something strictly better takes their slot.
How we choose models
- Weights are publicly downloadable without a sign-up wall (HuggingFace's basic gating is acceptable; opaque vendor licensing is not).
- The model is genuinely useful for at least one common task — we'd run it ourselves before recommending it.
- The licensing terms are clear enough that a non-lawyer can understand what they imply.
- The publishing organization has a stable enough release cadence that links won't rot within a year.
How we don't make money
There is no affiliate revenue on this site. Every "Download" link goes to the publisher's official mirror — Hugging Face, Ollama, Kaggle. We do not run sponsored placements, do not accept paid review slots, and do not boost models in the catalog in exchange for anything. The cookie banner you saw is for first-party analytics only; we don't ship third-party trackers.
Who maintains this
A rotating group of three or four contributors who happen to spend a lot of time fine-tuning and evaluating open-weight models. If you spot something wrong or outdated, the easiest way to reach us is to file an issue on the catalog's source repository.
What's intentionally missing
- Closed-weight commercial models. GPT-4, Claude, Gemini Pro, and their kin are not here. They're well-covered elsewhere and you can't run them locally anyway.
- Image, audio, and video models. This is a language model catalog. A separate site (KittyHawk Pixel Hub) covers diffusion models; it's currently a stub.
- Fine-tunes of catalogued models. Thousands of community variants exist for popular bases. We link to the most useful ones from each model's page rather than listing them separately.