tech
March 11, 2026
Search API: Better Extraction, Dynamic Benchmarks
Our Search API snippet quality has been optimized along two dimensions: relevance and size.
TL;DR
- Snippet quality optimized for relevance and size through new extraction, span-level labeling, and evaluation systems.
- Span-level labeling precisely evaluates snippet content, identifying vital and irrelevant segments.
- Improved content understanding pipeline handles a broader range of structured data, including tables and dynamic content.
- Smaller, more relevant snippets reduce response payload size and latency, leading to lower token costs for developers.
- Introduced SEAL benchmark to test time-sensitive retrieval capabilities.
- Search API now supports up to 5 queries per request, returning results grouped by query.
- Expanded filtering includes language (ISO 639-1) and regional (ISO country code) search, in addition to domain and recency filtering.
- Perplexity's Python SDK now offers native support for the Search API.
Continue reading the original article