Enhanced search model achieving best-in-class factuality with F-score of 0.858 on SimpleQA benchmark. Designed for complex multi-step tasks with double the citations of base Sonar, providing deeper understanding and context retention.
- Best-performing model on factuality benchmark with F-score of 0.858, leading the industry in accuracy
- Enhanced capabilities for complex multi-step queries with double the citations compared to base Sonar
- Advanced extensibility handling in-depth research with greater accuracy and comprehensive sourcing
- Designed for enterprise applications requiring deep understanding and sustained context retention
- Superior performance compared to search-enabled GPT-4o while maintaining significantly lower costs
Model Information
Supported Formats
SonarIn-house search-grounded model based on Llama 3.3 70B, optimized for speed and cost-efficiency with real-time web research capabilities. Ideal for straightforward factual queries requiring fast, cited responses with lightweight question-and-answer integration.
Sonar ReasoningChain-of-thought reasoning model with embedded live search capabilities, specialized for logical reasoning and step-by-step inference. Excels in mathematical, coding, and analytical challenges requiring sustained reasoning over multiple steps.
Sonar Reasoning ProAdvanced reasoning model powered by DeepSeek-R1 1776 integration, delivering superior step-by-step inference and comparison capabilities. Optimized for research synthesis, evaluation tasks, and complex analytical workflows requiring deep reasoning.