Perplexity uncensors DeepSeek-R1 with open-source R1 1776

R1 1776 is an open-source reasoning model from Perplexity, built to strip political censorship out of DeepSeek-R1 without dulling its analytical edge. It gives developers and researchers an uncensored alternative that delivers straight, factual answers on sensitive topics while matching the base model's high-end math and logic performance.

What it does

DeepSeek-R1 showed the world that open-weight models could trade punches with OpenAI's o1 and o3-mini in pure reasoning. However, the original release frequently dodged politically sensitive queries or fell back on state-aligned narratives.

R1 1776 takes aim directly at that limitation. According to Perplexity, the new model offers factual and neutral responses across historically restricted topics, all while running neck-and-neck with the original DeepSeek-R1 on core reasoning benchmarks.

How they built it

Perplexity didn't spend millions retraining the model from zero. Instead, they ran a specialized post-training pipeline using Nvidia's NeMo 2.0 framework.

The team first identified roughly 300 sensitive topics subject to censorship. They built a custom multilingual classifier to collect relevant prompts, assembling a clean dataset of 40,000 multilingual user queries gathered with consent and privacy protections in place.

Fine-tuning on this focused set allowed the model to drop its conversational guardrails without degrading its underlying chain-of-thought capabilities.

Where to get it

The model weights for R1 1776 are fully open and available to download right now on Hugging Face.

If you do not want to spin up local infrastructure, you can also access the model directly via Perplexity's Sonar API.

Why it matters

This is an essential release for researchers, journalists, and developers who need frontier-grade reasoning on world events and history without hitting arbitrary refusal walls.

Uncensored community fine-tunes often suffer from brain damage on technical benchmarks. Perplexity managed to break the ideological handcuffs while keeping DeepSeek's powerhouse reasoning intact.