R1 1776

Perplexity’s R1 1776 is a de-censored reasoning model derived from DeepSeek-R1 and released open-source (MIT). Perplexity identified ~300 CCP-censored topics, mined a multilingual corpus of ~40k user prompts, and post-trained the model—capturing valid chain-of-thought traces—using NVIDIA’s NeMo 2.0 framework. Internal evaluations report that R1 1776 maintains the base R1’s math/reasoning strength while answering previously suppressed queries directly. You can download weights from Hugging Face or call it through the Sonar API; community builds (e.g., Ollama) are also available. Use R1 1776 when you need R1-level reasoning without censorship on sensitive geopolitical topics.