What is the sonar-pro-high model in lmarena?
- User
What is the sonar-pro-high model in lmarena?
Assistant
Overview of the sonar-pro-high Model in LM Arena
sonar-pro-high (often referenced as
ppl-sonar-pro-high) is a search-augmented large language model (LLM) developed by Perplexity and evaluated in the LM Arena's Search Arena leaderboard. This model is part of the broader Sonar family, which is specifically optimized for deep web search, comprehensive reporting, and robust citation practices[7][5].Key Features and Performance
- Search-Augmented LLM: The sonar-pro-high model integrates real-time web search results into its responses, enabling it to provide up-to-date, well-sourced answers[7][5].
- Citation Depth: It is distinguished by its ability to cite 2–3 times more sources than comparable models, such as Google's Gemini or OpenAI's GPT-4o search variants[7][2][5].
- Leaderboard Ranking: In the LM Arena Search Arena evaluation,
ppl-sonar-pro-highconsistently ranked among the top models. According to the April 2025 leaderboard, it achieved an Arena Score of 1071 (with a 95% confidence interval of +15/-10) and was grouped with other top-performing Sonar models, just below the reasoning-augmented variants and Google's Gemini-2.5-Pro-Grounding[7]. - User Preference: The model is favored for producing longer, more detailed responses with extensive citations, which aligns with user preferences for transparency and depth in AI-generated answers[7][5].
- Availability: sonar-pro-high is accessible to Perplexity Pro users and via the Sonar API, making it available for both end users and enterprise integration[5].
Comparison Table: Top Models in LM Arena Search Arena
Model Arena Score 95% CI Organization gemini-2.5-pro-grounding 1142 +14/-17 Google ppl-sonar-reasoning-pro-high 1136 +21/-19 Perplexity ppl-sonar-reasoning 1097 +11/-17 Perplexity ppl-sonar 1072 +15/-17 Perplexity ppl-sonar-pro-high 1071 +15/-10 Perplexity ppl-sonar-pro 1066 +12/-13 Perplexity gemini-2.0-flash-grounding 1028 +16/-16 Google api-gpt-4o-search 1000 +13/-19 OpenAI api-gpt-4o-search-high 999 +13/-14 OpenAI [7]
Distinction from Other Sonar Models
- sonar-pro-high is a non-reasoning variant, meaning it does not include the advanced reasoning capabilities found in the "reasoning" versions (e.g.,
ppl-sonar-reasoning-pro-high), but it still delivers strong performance in search and citation tasks[7]. - It is optimized for users and applications that prioritize comprehensive, well-sourced information over complex multi-step reasoning.
Summary
The sonar-pro-high model in LM Arena is a top-performing, search-augmented LLM from Perplexity, recognized for its deep citation practices and comprehensive answers. It ranks just below the reasoning-augmented models and Google's top search model, making it a leading choice for users seeking detailed, well-sourced AI-generated content[7][5][2].
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.