Perplexity adds web grounding to its Decisions API through Fast Search
Article timeline
The articles in this story. Times are ET.
Aravind SrinivasOfficialPerplexity integrates Fast Search into its Decisions API for web-grounded answersAIPerplexity says it is integrating Fast Search into its Decisions API so batch-mode decisions can draw on web facts. Search costs $0.001 per request, and the Decisions API costs $0.02 per 1M input tokens.
Perplexity DevelopersOfficialPerplexity describes uses for web grounding in fact-checking and monitoringAIPerplexity says web grounding can fact-check a list of claims, check whether a pinned package has a newer stable release, and flag accounts with a new CEO, an acquisition, or layoffs. The post links to a web-grounded cookbook with further guidance.
- Perplexity DevelopersOfficialPerplexity explains using Fast Search to ground answers in web pages
AIPerplexity says that when Fast Search is enabled, its results join the state as evidence before the decision model answers. Passing a URL as the state lets users ask about a specific page. Requests cost $0.001 per search and $0.02 per 1M tokens.
- Perplexity DevelopersOfficialPerplexity's pplx-decider-v1.2 scores 86% on PWS-Decisions with web grounding
AIPerplexity says pplx-decider-v1.2 with web grounding scores 86% on 100 public PWS-Decisions questions, up from 51% without it. The reported median latency was 370 ms, with p95 at 685 ms.
- Perplexity DevelopersOfficialPerplexity adds web grounding to its Decisions API
AIPerplexity says the Decisions API can now search the web before answering, using "mode": "web_grounding". With this mode, Fast Search runs one query per question, so the pplx-decider-v1.2 model answers from current web information rather than only its training data and the state provided.
