Benchmarks & research
FLock.io's this-that-model-1.0 is a 1.88B model that picks one of the options a program declares and gives each a probability, without generating a single token.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
It was adapted from Mapika's decider-2b and comes with an arXiv paper and a spatial benchmark. Because the answer is read only over the declared options, an answer outside the list cannot occur. The authors flag that their benchmark's question shapes were seen in training, and that multi-step arithmetic remains a clear weakness.
It suits a program that keeps reaching the same kind of judgement call, such as whether a refund is within policy or a shell command is safe to run unattended. The model picks one listed answer and reports how sure it is, so a low number can send the case to a person or a bigger model instead.