Benchmarks & research
Darwin-27B-ZTC-v2 is VIDRAFT's open Apache-2.0 decision model that answers yes/no, choice and score questions in a single pass, without writing any text.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
On 10 October it led the S1MB leaderboard's default ranking of 102 models, with Jev 1.13 fifth; counting only the board's generalization tasks, Jev was first and Darwin third. Darwin was trained on Open-Jev training data, and 22 S1MB benchmarks come from the same tasks' test sets, although the authors report finding no exact overlap.
Darwin picks between answers you supply and says how likely each one is, such as whether support should refund a double charge, pass it to billing or close it. It runs on your own large graphics card. Its first place does not hold on the leaderboard's tasks that test unfamiliar kinds of work.