Vote for the next model to test

Search for a model below. If someone already asked for it, give it a +1. Not there yet? Request it with a Hugging Face link or id and it joins the list. Requests are checked against Hugging Face, so only real, public models make it in.

Most votes go first. A model with one vote is tested oldest first while it is under two days old; older single-vote requests wait until someone else votes for them.

We run everything on our own hardware, so we can’t promise a date, but we do read the list. Models too big for our rig are never turned down for their size: they wait on the Needs funding tab.

Find a model

Type a few letters to see what others asked for, then press +1 on the one you want.

Up to about 135 billion parameters fits our rig (Q8_0, then Q6, then Q5, never lower). Anything bigger goes to Needs funding, or waits there until a hosted service runs it. What the rig can test

The id from the model's Hugging Face page, e.g. Qwen/Qwen3-32B. Pasting the page link works too, and so does a model card or a Hugging Face link inside other text. If we find the original model behind a quantized copy, we'll say so. If someone already asked for it, your request counts as a +1.

Loading the list…

Send feedback

Spotted a mistake on the board, want a model that isn't on Hugging Face (a hosted API, say), or have something else to add? We read all of it.

Searching, voting and sending run a short automatic check in your browser (a few hundred milliseconds of arithmetic, no puzzle to solve) to slow down scripts; it does not identify you. The first time it passes, your browser keeps a random pass for 30 days, tied to your network address, not to your name or email. What's stored for a request: the model id, its vote count and a salted hash of your network address, so one network counts once, never the address itself. Feedback is stored on the server for us to read, with the optional email for up to 180 days after we have handled it, and is never shown on the site. Privacy policy.