Cette section n'existe pas encore en français — le contenu ci-dessous est en anglais. La traduction arrive.
Sonar Pro
Perplexity's flagship search-grounded model, optimized for real-time web retrieval, citation accuracy, and synthesis.
Quality
Modality
text
Context
127K tokens
Access
closed
Fabian's Take
"When I need answers grounded in real-time facts rather than model pre-training, Sonar Pro is my default. It powers Perplexity's Pro search, and its true superpower is how it translates natural language questions into optimized search queries and then cleanly summarizes the findings with proper citations."
Sonar Pro is Perplexity’s premier proprietary model, custom-tuned specifically for search and retrieval-augmented generation (RAG) tasks. Rather than relying solely on static training data, Sonar Pro acts as an agent that interactively queries the web, reads multiple sources, and compiles them into a coherent answer.
Real-Time Search Native
Standard models struggle with RAG because they are not trained to read search snippets or handle conflicting search results. Sonar Pro is fine-tuned from the ground up to analyze search results, filter out SEO spam, and synthesize information from high-quality sources.
Structured Citations
The model is exceptionally disciplined at citation mapping. Every factual claim is coupled with an inline reference to the source URL, which dramatically reduces hallucination rates compared to generic text models performing manual search prompts.
The Verdict
Best for: Real-time web research, fact-checking, and synthesizing multiple online sources with inline citations.
Pros
- Industry-leading citation accuracy and source grounding
- Extremely fast response times for web-search questions
- Excellent at synthesizing complex, changing news topics
Cons
- Weaker than Claude or GPT-o3 for complex multi-step programming or creative writing
Specs
- Pricing $1.00/M tokens (+ $5.00/K search requests)
- Cost Tier moderate
- ⚡Speed Tier fast
- License Proprietary
