Sonar API

Perplexity's developer API. Gives programmatic access to its search-grounded LLMs. Four tiers: Sonar (lightweight factual), Sonar Pro (multi-source complex queries), Sonar Reasoning Pro (analytical with chain-of-thought), and a Deep Research endpoint. In 2026 the platform also resells OpenAI, Anthropic, Google and xAI models at provider rates plus 0.005 dollars per web search call.

• Four Sonar tiers covering quick lookups to multi-step research
• Citations returned as structured fields, not buried in prose
• Now a multi-provider marketplace: OpenAI, Anthropic, Google, xAI all callable through one key
• Pay-as-you-go pricing with no monthly minimum
• Designed for grounded answers, not free-form generation

• Build an internal Q&A tool that must always cite sources
• Power a compliance assistant that needs to ground every answer in policy documents and live web sources
• Add a sourced search box to a customer-facing knowledge base
• Replace a bare LLM call in an agent workflow where hallucination is the main risk

More expensive per token than calling OpenAI or Anthropic directly. The search-grounding fee adds up at scale. Latency is higher than a plain LLM call because it actually goes to the web. Documentation is decent but not as mature as OpenAI or Anthropic's.

Sonar at 1 dollar per million input and output tokens; Sonar Pro at 3 dollars input, 15 dollars output per million tokens; Sonar Reasoning Pro at 2 dollars input, 8 dollars output. Plus 5 to 14 dollars per 1,000 requests depending on tier.