Benchmarks & research
Valen looks at text, a photo or a video clip and answers questions such as which animal is in the picture, with a probability for every option you offer.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
Valen-Team released 0.8B, 2B and 4B versions on 7 October, built on Qwen3.5 with a small decision head, and the README replays the model solving Sokoban and other grid games step by step. An earlier September preview differs from these releases, and every score so far is the maker's own.
Valen suits questions about pictures or short videos, such as which animal is in a photo or whether a statement about the scene is true. You can ask several questions about the same clip at once, and it reads the clip only once to answer them all. Setting it up takes some Python and a computer with a graphics card.