R1 1776 is a post-trained version of the DeepSeek-R1 large language model, released by Perplexity AI with a focus on uncensored, factual, and high-reasoning output. It was specifically fine-tuned to remove censorship associated with the Chinese Communist Party while preserving strong reasoning abilities. The model was evaluated using a multilingual dataset covering over 1,000 sensitive topics, ensuring open engagement across controversial subjects. Human annotators and LLM-based evaluators confirmed that the decensoring process did not compromise performance. R1 1776 maintains parity with the original R1 model in benchmarks for math and reasoning. Licensed under MIT, the model is accessible for both research and commercial purposes. It is positioned as a transparent, high-performing alternative for users seeking unrestricted conversational AI.
Features
- Based on DeepSeek-R1 with 671B parameters
- Fine-tuned to eliminate CCP-related censorship
- Maintains advanced reasoning and math capabilities
- Evaluated on a multilingual, sensitive-topic benchmark
- Human and automated evaluations confirm decensoring quality
- Released under the permissive MIT license
- Supports text generation and conversational use
- Available via Hugging Face for research or commercial use
Project Samples
