Perplexity Launches Batch API on Agent API With 50,000 Requests Per File
Perplexity released a Batch API on its Agent API, allowing users to upload a JSONL file of up to 50,000 requests, create a batch, and download results when finished. Each batch request runs the full Agent API loop, so built-in tools such as web_search and fetch_url work inside a batch, and any model from the Agent API models list can be used. A cookbook tutorial demonstrates enriching a CSV of companies overnight with one web-grounded request per row.
Event version 1.
Key insights
Batch size and workflow
The Batch API accepts a JSONL file of up to 50,000 requests. Users create a batch and download the results when it finishes.
Full agent loop in batch mode
Every batch request runs the full Agent API loop, so built-in tools including web_search and fetch_url work inside a batch. Any model from the Agent API models list can be used.
Data enrichment tutorial
A new cookbook tutorial enriches a CSV of companies overnight. It runs one web-grounded Agent API request per row, writes to JSONL, runs as a batch, and joins results back to the rows as enriched data.
Scale example
The batch capability supports data enrichment workloads at scale, such as a spreadsheet of 50,000 companies with live web data overnight.