Perplexity

Recently shipped native document input and image search for the Agent API.

The Agent API now reads documents and images.

• Native document input takes PDF, DOC, DOCX, TXT, and RTF as input_file content, inline or from a public HTTPS URL. Multi-turn conversations continue with previous_response_id without re-uploading inline documents.
• image_search returns typed image_search_results with image and source-page URLs, plus result limits and domain, format, and safe-search filters. It costs $2.50 per 1,000 successful invocations.

Search presets got cheaper and faster. The low, medium, and high presets now use Fast Search for web_search, cutting cost from $2.50 to $1.00 per 1,000 invocations and saving roughly 800 ms. xhigh is unchanged.

Sandbox extraction now reports what it costs. Token usage and cost for successful pplx_sdk.llm.extract calls show up in usage.tool_calls_details.sandbox and roll into usage.cost.total_cost, priced with openai/gpt-6-luna. Historical responses are not backfilled.

Decisions API switched its default model. pplx-decider-v1.1-27b replaces pplx-decider-v1-27b, with input at $0.02 per million tokens, down from $0.04. The old name still works at the same rate. Claude Haiku 5.5 (anthropic/claude-haiku-5-5) also joined the Agent API; requests over 100k input tokens use higher rates.