Typed Decisions Leaderboard: A New #1 at 0.81
aibenchmarksevaluationmachinelearning
TL;DR
The Hugging Face Typed Decisions official benchmark has a new #1: Darwin-27B-ZTC-v5 at 0.81, holding both first and second place, ahead of the next entries at 0.74.
Disclosure: Darwin-27B-ZTC-v5 is from VIDRAFT, a technology partner of QuantID.
What is being measured
Typed Decisions is a calibrated, structured decision task: a model chooses among named options under stated constraints and reports a confidence. It is scored on a public Hugging Face leaderboard built from the eval results that model repositories publish, so the ranking reflects the public record.
Reading the gap
A 0.07 margin to #2 is meaningful on this board, and taking both #1 and #2 means the result is not a single lucky checkpoint. For anyone evaluating decision models, the useful questions are the margin to #2 and whether the same system shows up more than once near the top. Both hold here.
Leaderboard: https://huggingface.co/datasets/LocalLLaMA/typed-decisions