Sonar Models by Perplexity Labs
Details
Capabilities
Links
WebsiteAbout
The Sonar family of Large Language Models (LLMs), developed by Perplexity AI, marks a notable progression in terms of cost-efficiency, speed, and performance compared to predecessors like PPLX and Mixtral 5. These models are distinct for their ability to provide real-time internet access and deliver up-to-date information, which is a significant improvement over traditional LLMs 5. The Sonar family comprises different model configurations, such as "small" and "large" models, each tailored for specific tasks and featuring varying context window lengths 67. Built upon the foundation of the Llama 3.1 model, Sonar models are further refined with Perplexity's proprietary search capabilities to enhance accuracy and relevance 10. They offer both online and chat versions, catering to a wide range of applications that require either rapid responses or more extended conversational interactions 57. Additionally, these models are available through APIs, facilitating their integration into various applications seamlessly 67.
Current Variants
Use-when guidance is derived from seed capabilities, context, release, and replacement fields.
Use when the workload needs 200k context and structured outputs.
Use when the workload needs 128k context and structured outputs.
Use when the workload needs 128k context and structured outputs.
Use when the workload needs 127k context and structured outputs.
Use when the workload needs 200k context and structured outputs.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Sonar Pro Search | Use when the workload needs 200k context and structured outputs. | 2025-11 | 200k contextstructured outputs | Current |
| Sonar Deep Research | Use when the workload needs 128k context and structured outputs. | 2025-03 | 128k contextstructured outputs | Current |
| Sonar Reasoning Pro | Use when the workload needs 128k context and structured outputs. | 2025-03 | 128k contextstructured outputs | Current |
| Sonar | Use when the workload needs 127k context and structured outputs. | 2025-01 | 127k contextstructured outputs | Current |
| Sonar Pro | Use when the workload needs 200k context and structured outputs. | 2025-01 | 200k contextstructured outputs | Current |
Release Timeline
3 release groupsReplaced By
Keep for legacy integrations; evaluate Sonar Reasoning Pro before new work.
Specifications(6 models)
| Model | Released | Context | Structured Outputs |
|---|---|---|---|
| Sonar Pro Search | 2025-11 | 200k | Yes |
| Sonar Deep Research | 2025-03 | 128k | Yes |
| Sonar Reasoning Pro | 2025-03 | 128k | Yes |
| Sonar | 2025-01 | 127k | Yes |
| Sonar Pro | 2025-01 | 200k | Yes |
Available From(3 providers)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| Sonar | OpenRouter | $1 | $1 | Serverless |
| Sonar | Perplexity Labs | $1 | $1 | Serverless |
| Sonar Reasoning Pro | OpenRouter | $2 | $8 | Serverless |
| Sonar Deep Research | OpenRouter | $2 | $8 | Serverless |
| Sonar Reasoning Pro | Perplexity Labs | $2 | $8 | Serverless |
| Sonar Deep Research | Perplexity Labs | $2 | $8 | Serverless |
| Sonar Pro | OpenRouter | $3 | $15 | Serverless |
| Sonar Pro Search | OpenRouter | $3 | $15 | Serverless |
| Sonar Pro | Perplexity Labs | $3 | $15 | Serverless |
| Sonar Pro Search | Perplexity Labs | $3 | $15 | Serverless |
Frequently Asked Questions
- What is Sonar used for?
- Sonar is used for structured outputs. The family description and listed model capabilities point to those workloads as the best fit.
- How does Sonar compare to Claude Fable?
- Sonar by Perplexity Labs is strongest where you need structured outputs, while Claude Fable by Anthropic is the closest related family to check for vision and multimodal work. Sonar has 6 listed variants and reaches up to 200k context, while Claude Fable reaches up to 1m context, so compare the specs and pricing tables before choosing a production model.
- Which Sonar model should I use?
- For the lowest listed input price, start with Sonar through Perplexity Labs at $1/1M input tokens. For the most capable/latest local choice, evaluate Sonar Pro Search with 200k context and structured outputs.
