About Home AI Bench
What it is
Home AI Bench measures how fast open image models run on ordinary home GPUs and shows what they produce, side by side, with the failures included. It exists to help people with 8 to 12 GB cards decide what to try, and to keep a public record of what worked and what did not.
Independence
- No advertising and no accounts. Analytics only run if you accept them.
- Nobody pays for results. We are not affiliated with the makers of the models or with the companies that sell the hardware we test on. If we ever accept support, such as free compute credits, we will say so on this page and in the notes it touches.
- Models are chosen because people ask about them or because they are new and open. We say when a license limits use, and we keep our tests within each license.
How we work
Methods, settings and limits are on Methods. Every claim in a note comes from a run we can name. Drafts and commentary may be written with the help of AI tools; numbers, settings and sources are checked by a person before anything is published.
Corrections
If a published statement turns out to be wrong, we fix the page and add a visible correction that says what was wrong and what changed. An example is the signature correction in the MOSS-Image note.
Images and rights
All pictures are generated by us and carry a watermark. If a picture looks like a real person, copies someone's work or shows a name or signature, please report it and we will take it down while we check.
Follow and contact
Follow @homeaibench on X or @homeaibench.bsky.social on Bluesky, subscribe to the RSS feed, or use the report form.