Benchmarks & research

Qwen3.5-0.8B AA Decision

This is a compact 0.8-billion-parameter decision model fine-tuned from Qwen3.5 to answer structured questions using the System-One API.

Bookmark Keep this in your collection.

Leave a noteWhat would you try with this?

Only you can see your notes.

Source screenshot of Qwen3.5-0.8B AA Decision — SOURCE SCREENSHOTFull screenshot ↗

What it does

The creator reports requests taking about 68 milliseconds on a consumer graphics card, but the small model has limited reasoning. Results are near chance on natural-language inference tasks such as ANLI, and complex 77-way classifications are a struggle compared with larger models.

How you can use it

Use llama.cpp to run the downloadable model with a compatible decision server, then supply a situation and fixed answers to choose or rate. Alok reports a compressed version fitting in under 4 GB of graphics memory. Its small training set and weak reasoning results make tests on your particular task important.