Parallel Launches Decisions API Combining Live Web Retrieval With Jev-Style Primitives
Parallel introduced a web-grounded Decisions API on October 25 that retrieves live evidence and returns shared citations alongside noul, choice, and score answers.
Parallel Web Systems expanded the decision-model category in a new direction on October 25 by launching the Parallel Decisions API—an endpoint that adopts TypeSafe’s noul, choice, and score question primitives but searches the live web before answering.
Built for decisions whose evidence lives outside your database
Unlike token-priced decision models such as Jev or OpenAI’s Decisions API, which reason solely over text or images passed in the request, Parallel’s endpoint accepts a seed state (such as a company name, URL, or JSON record) and up to 100 named questions. Parallel retrieves a shared set of current web pages, evaluates every question against that evidence, and returns typed answers alongside request-level citations containing URLs and supporting excerpts—typically in under a second.
Different economics for different workflows
Parallel prices the service at $1 per 1,000 requests, search included. As Parallel’s own documentation notes, that makes it more expensive than Jev ($0.042 per million input tokens) whenever the answer is already inside your own text, support ticket, or transcript. Where it competes is vendor verification, KYB rechecks, lead qualification, and claim checking—workflows where teams would otherwise stitch together a separate search API and an LLM call.