Benchmarks & research

StartLux-Decision

StartLux Labs' open decision models, from 0.8B to 27B plus a 35B mixture of experts, answer Jev-format questions about text and images with a probability per option.

Bookmark Keep this in your collection.

Leave a noteWhat would you try with this?

Only you can see your notes.

Source screenshot of StartLux-Decision — SOURCE SCREENSHOTFull screenshot ↗

What it does

Its demos show the 27B model making the choices in recorded runs of chess against Jev 1.13, Super Mario Bros., Doom, StarCraft II and a mock web shop, while game or harness code does the aiming, building and clicking. The code is Apache-2.0, but the weights are CC BY-NC 4.0, so commercial use needs a separate licence.

How you can use it

StartLux-Decision fits research and hobby projects that want Jev-style answers from a model they run themselves, such as picking a game move or sorting support messages. A business would need a separate licence from StartLux first. Its benchmark wins come from the makers' own runs, so check it on your own examples.