Perplexity’s Hugging Face repository presents pplx-decider-v1.1-27b as an update to pplx-decider-v1-27b. Observed facts: the repository reports a Decision Index increase from 56.4 to 61.56, and says the model is more than 3.5 points ahead of Jev. It attributes most of the gain to lifting the causal mask and training on more data.
Mechanism: The stated changes alter the model’s training setup and data exposure; the source links those changes to a higher Decision Index, but does not isolate their individual effects. Why now: This release provides a fresh, quantified comparison for teams tracking models built for decision tasks.
First implication: Builders can treat the new checkpoint and score as a candidate for their own evaluation, rather than as proof of improved performance across decision workloads. Evidence boundary: The supplied repository excerpt does not specify the benchmark methodology, test composition, or independent replication, so the reported score and comparison remain claims from the model’s source.
First watch condition: Whether Perplexity or independent evaluators publish enough benchmark detail and replication results to test the reported gain. That evidence would clarify whether the change generalizes beyond the stated index.