Perplexity open-sources r1-1776, a uncensored DeepSeek r1 variant
Feb 18, 2025
Key Points
- Perplexity open-sourced r1-1776, a DeepSeek r1 variant stripped of content censorship while maintaining math and reasoning benchmark performance.
- The move addresses Western developer concerns about hidden constraints in DeepSeek models by publicly demonstrating uncensoring can preserve core capabilities.
- Perplexity ranks 26th in productivity apps despite technical differentiation, suggesting tactical product moves have not yet overcome entrenched switching costs to Google and ChatGPT.
Summary
Perplexity open-sources r1-1776, an uncensored DeepSeek r1 variant
Perplexity has open-sourced r1-1776, a version of DeepSeek r1 that has been post-trained to remove censorship on sensitive topics while preserving the model's math and reasoning capabilities. The name—a deliberate Americana reference—signals the company's positioning around unrestricted information access.
To validate the uncensoring work, Perplexity created a diverse, multilingual evaluation set of 1,000 examples using human annotators and LLM judges. They then compared censorship frequency in the original r1 against state-of-the-art models and their modified version. Benchmark evaluations showed r1-1776 performed on par with the base r1 model, indicating that removing censorship constraints had no degradation impact on core reasoning and math abilities.
The move addresses a real concern in the timeline: whether DeepSeek's models ship with hidden biases or constraints that make them less useful for Western developers and enterprises. By open-sourcing an explicitly uncensored variant, Perplexity surfaces the work required to strip those constraints and demonstrates it can be done without sacrificing reasoning quality.
Product strategy and market positioning
Perplexity's broader product move—bundling r1-1776 with their own deep research competitor—fits a reactive pattern. The company has been quick to follow competitors' launches: after OpenAI released deep research, Perplexity launched their own; after Grok three's announcement, they unveiled r1-1776. The CEO frequently uses responses to Sam Altman's posts as marketing hooks, keeping Perplexity visible in the competitive narrative.
The company does offer differentiated features. Their algorithmic feed of AI-generated news summaries with source attribution is well-executed, and their search product integrates reasonably cleanly. Yet app store rankings tell a harder story: Perplexity sits at number 26 in the productivity category, behind Grok (2), ChatGPT (1), and DeepSeek (3). More significantly, it ranks below unrelated apps like HP Smart and various VPN tools. Consumer adoption momentum, despite the technical work, remains limited.
The tension is between novelty and friction. Users already default to Google search and ChatGPT because switching costs—both cognitive and habitual—are real. For Perplexity to move meaningfully up the rankings, the product would need to be either substantially faster than Google or so obviously superior that users reroute their Chrome default search. Tactical marketing moments, however well-executed, have not yet translated into that kind of traction.