Models designed to retrieve and synthesize information efficiently.
sonar-pro
Advanced search offering with grounding, supporting complex queries and follow-ups. Learn more →
sonar
Lightweight, cost-effective search model with grounding. Learn more →
Best suited for quick factual queries, topic summaries, product comparisons, and current events where simple information retrieval and synthesis is needed without complex reasoning.
Not ideal for multi-step analyses, exhaustive research on broad topics, or projects requiring detailed instructions or comprehensive reports across multiple sources.
Models that conduct in-depth analysis and generate detailed reports.
sonar-deep-research
Expert-level research model conducting exhaustive searches and generating comprehensive reports. Learn more →
Ideal for comprehensive topic reports, in-depth analysis with exhaustive web research, and projects requiring synthesis of multiple information sources into cohesive reports like market analyses or literature reviews.
Avoid using for quick queries, simple lookups, or time-sensitive tasks, as these models may take 30+ minutes to process and are excessive when depth isn’t required.
Premier reasoning offering powered by DeepSeek R1 with Chain of Thought (CoT). Learn more →
sonar-reasoning
Fast, real-time reasoning model designed for quick problem-solving with search. Learn more →
Excellent for complex analyses requiring step-by-step thinking, tasks needing strict adherence to instructions, information synthesis across sources, and logical problem-solving that demands informed recommendations.
Not recommended for simple factual queries, basic information retrieval, exhaustive research projects (use Research models instead), or when speed takes priority over reasoning quality.
A version of DeepSeek R1 post-trained for uncensored, unbiased, and factual information. Learn more →
Perfect for creative content generation, tasks not requiring up-to-date web information, scenarios favoring traditional LLM techniques, and maintaining conversation context without search interference.
Unsuitable for queries needing current web information, tasks benefiting from search-augmented generation, or research projects requiring integration of multiple external sources.