R1 1776
DeepSeek-R1 post-trained to remove censorship
About
R1 1776 is Perplexity's post-trained version of DeepSeek-R1, created to address the base model's refusal to answer politically sensitive topics. Perplexity identified around 300 censored topics, built a 40,000-prompt multilingual dataset, and fine-tuned with NVIDIA's NeMo 2.0 framework. The company reports reasoning performance on par with the original R1. The weights are MIT-licensed on Hugging Face and it is also served through Perplexity's Sonar API. It inherits R1's architecture and limits — it is still a large MoE that needs serious hardware to self-host, and "uncensored" reflects Perplexity's post-training choices rather than the absence of any safety tuning.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Perplexity-reported: on par with base DeepSeek-R1 on math and reasoning benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Perplexity announcement Feb 18 2025 confirmed across multiple outlets: post-trained DeepSeek-R1, 40K multilingual prompts, NeMo 2.0.; Reasoning on par with base R1 per Perplexity evaluations.; License: MIT weights on Hugging Face (perplexity-ai/r1-1776) per reporting.; Limitation: inherits R1's hardware demands; 'uncensored' reflects Perplexity's tuning choices..