Every week another open-weight model claims to beat GPT or Claude. We track what's actually real — what you can download, what hardware it needs, and how it performs — and put it next to the closed models everyone already knows, so you can judge for yourself.
You can download these and run them yourself — on your own computer, or on any cloud service you choose. Once you have the files, no company can take them away, shut them down, or change how they work.
These only run on the maker's own servers. You use them through an app or a paid connection — like ChatGPT or Claude — but you can never download them or run them yourself.
Open and closed models come out of very different ecosystems — American labs, Chinese labs, European labs, sometimes a single developer — and each publishes its own release notes, benchmark chart and licence page. AISNAK pulls it all into one place, so a closed frontier model and an open one released last week sit side by side, and you're not the one stitching ten tabs together to compare them.
Six tools, built for the six questions everyone asks once they start looking.
Find a model by name, family or what you actually need it for — filtered live as you type, across open and closed alike.
Search models →Filter by RAM and see the GPU a model actually needs, before you spend an evening downloading something that won't run.
Check hardware →Line up to fifteen models at once — parameters, context length, licence, benchmark scores — in one scrollable table.
Start a comparison →No GPU at home? Rent one by the hour. We've compared providers on price and how long it takes to get running.
Compare providers →The same tests, run the same way, across every model — so a score means the same thing wherever it came from.
View benchmarks →The top-ranked models right now, open and closed together, so you can see who's actually out in front.
See the leaders →Every model in this directory is available as a free JSON API — no sign-up needed to read it.