Perplexity open-sources R1 1776, DeepSeek R1 uncensored model

Perplexity has released an open-source version of the DeepSeek R1 model, named R1 1776, which has been adjusted to deliver uncensored, unbiased, and factual responses. The model was evaluated using a diverse, multilingual set of over 1000 examples to ensure it maintained its reasoning capabilities while reducing censorship.

Evaluations across multiple benchmarks showed that the post-trained model performed as well as the base R1 model, indicating that the de-censoring did not affect its core reasoning capabilities.