Hugging Face released Darwin-27B-ZTC-v2 on October 9, 2026, saying it is a zero-token classifier that emits probability distributions directly. It is the company's first System One model update since the introduction of the System One Mosaic Benchmark.
Hugging Face reported a Borda score of 89.58, measured on the System One Mosaic Benchmark. That compares with a score of 87.50 for OpenJev-27B, the second-ranked model.
Darwin-27B-ZTC-v2 is built on a structured decision-making framework and targets operational tasks such as routing, grading, and gating. Availability begins with public leaderboard access, initially for researchers and developers.
"A single model that must cover all three cannot overfit to one decision geometry," said the blog post. The System One Mosaic Benchmark was built to measure exactly these models.
The announcement follows the launch of the System One Mosaic Benchmark. Hugging Face framed the significance as a step toward measuring models that make typed decisions rather than generate free text.
Hugging Face did not say how the model performs on open-ended generation tasks, and the source raises the open question of whether the model is suitable for use cases beyond typed decisions. The source says the model remains high on generalization splits.
Source: huggingface
