Improved Sonar Models: Industry Leading Performance at Lower Costs

We're excited to announce significant improvements to our Sonar models that deliver what matters most to you: superior performance at lower costs. Our latest benchmark testing confirms that Sonar and Sonar Pro now outperform leading competitors, like search enabled GPT-4o, while maintaining significantly more affordable pricing. Through comprehensive testing and user feedback, we've developed a tiered approach that allows you to balance performance and cost efficiency without sacrificing the factual accuracy that makes Sonar stand out.

The improvements include a simplified billing structure, enhanced search capabilities, and the flexibility to optimize for both cost and performance based on your specific needs.

What this looks like at a glance:

• Three new search modes: Choose between High, Medium, and Low modes to match your needs for pricing and performance across all models except Sonar Deep Research

• Simplified billing: Transparent pricing for input/output tokens and search modes across models. For Sonar Pro and Sonar Reasoning Pro users, we will not charge for citations tokens amongst responses in the new modes

New Search Modes for Greater Performance and Price Control

Based on how customers use our API, we recognized that different use cases require different levels of search depth and context.

Adding Additional Modes for Sonar

To give you enhanced performance with better cost control, we've created three straightforward search options for Sonar:

• High: Maximum depth and context for handling more complex queries.

• Medium: Balanced approach for moderately complex questions.

• Low: Optimized for cost efficiency while maintaining strong accuracy for straightforward queries.*

* Low mode is equivalent to the current pricing for Sonar.

All options maintain Sonar's high standards for factual accuracy and response quality—varying only in the amount of search compute allocated based on the question's complexity.

These additional modes allow Sonar users to access customized performance without having to upgrade to Sonar Pro for a performance bump.

Better Performance at a Lower Cost for Sonar Pro

Similar to the base Sonar API, Sonar Pro and Sonar Reasoning Pro now have three search modes, with each tier providing enhanced capabilities for handling complex, multi-step queries with greater accuracy and affordability.

We've also made the decision to stop charging for citation tokens in responses across all search modes and models (except Sonar Deep Research). This change simplifies your billing and reduces costs while encouraging more comprehensive sourcing. Your applications will continue to receive the same high-quality, well-cited responses, but now at a more predictable price point.

Enhanced Performance and Cost Control

Our extensive benchmark testing demonstrates that Sonar outperforms competitors across key performance metrics while maintaining exceptional cost efficiency—a powerful combination that delivers unmatched value to developers and enterprises.

Benchmark uses the “High” Versions of the models

Testing reveals that Sonar achieves impressive factual accuracy that exceeds GPT-4o Mini and approaches GPT-4o High, while Sonar Pro surpasses even the most expensive competitor models. Most importantly, these performance gains come at a significantly lower price point, with standard Sonar costing a fraction of competitive offerings and Sonar Pro delivering premium capabilities at substantially greater value.

Our benchmarking confirms that Sonar and Sonar Pro occupy the sweet spot on the price-performance curve—delivering frontier-level capabilities without frontier-level pricing. With our new tiered approach, you can now select the perfect balance between depth, accuracy and cost for each specific use case.

Pricing Changes

To ensure a seamless transition, we'll support the current billing structure as the default option for the next 30 days. During this period, the new search modes will be available as opt-in features, allowing you to test and integrate them at your own pace. After this period, we'll switch to a default low context size option, though you'll still be able to select your preferred mode based on your specific requirements.

Important Note: After April 18, 2025, Sonar Pro and Sonar Reasoning Pro will not return Citation tokens or number of search results in the usage field in the API response

Use Sonar Today

Sonar excels at providing fast and accurate answers, making it a great model for everyday use. Perplexity Pro users can make Sonar their default model in their settings. It can also be used via the Sonar API offering.