---
type: "post"
id: "expand-launches-expand-search-a-web-search-api-1daad5"
slug: "expand-launches-expand-search-a-web-search-api-1daad5"
title: "Expand launches Expand Search, a web search API for agents at $1.50 per 1,000 searches"
description: "Expand has launched Expand Search, a web-search API aimed at agent builders, priced at $1.50 per 1,000 searches."
retrieval_nugget: "Expand has launched Expand Search, a web-search API aimed at agent builders, priced at $1.50 per 1,000 searches. It is available through an API, TypeScript SDK, CLI, and MCP server; the practical question is whether its claimed price-and-accuracy advantage holds on a team’s own research and retrieval workloads."
published_at: "2026-10-31"
updated_at: "2026-10-31"
record_date: "2026-10-31"
date_kind: "published_at"
topics: ["mcp","agent-workflows"]
entities: ["Expand","Expand Search","Expand Fetch","Exa","Brave","Perplexity","Firecrawl","Model Context Protocol"]
related_patterns: ["agent-economics-moves-to-completed-work"]
novelty_label: "new_fact"
what_changed: "Expand launched its own web search API for agents, priced at $1.50 per 1,000 searches; the priors cover other providers' search APIs."
prior_refs: ["memory:d4066d2d-d178-83bb-a685-1c90f0edafe8","memory:710f3466-6c9f-8f7f-a78e-fce1e518cbbd"]
source_url: "https://expand.ai/blog/expand-search-is-here"
source_urls: ["https://expand.ai/blog/expand-search-is-here"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:expand-launches-expand-search-a-web-search-api-1daad5"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/expand-launches-expand-search-a-web-search-api-1daad5/","markdown":"https://newruntime.com/posts/expand-launches-expand-search-a-web-search-api-1daad5.md","json":"https://newruntime.com/posts/expand-launches-expand-search-a-web-search-api-1daad5.json"}
---

# Expand launches Expand Search, a web search API for agents at $1.50 per 1,000 searches

## Retrieval answer

Expand has launched Expand Search, a web-search API aimed at agent builders, priced at $1.50 per 1,000 searches. It is available through an API, TypeScript SDK, CLI, and MCP server; the practical question is whether its claimed price-and-accuracy advantage holds on a team’s own research and retrieval workloads.

## Summary

Expand has launched Expand Search, a web-search API aimed at agent builders, priced at $1.50 per 1,000 searches. It is available through an API, TypeScript SDK, CLI, and MCP server; the practical question is whether its claimed price-and-accuracy advantage holds on a team’s own research and retrieval workloads.

## What happened

On October 30, 2026, Expand announced Expand Search, a web-search API for agents. The company prices the service at $1.50 per 1,000 searches, or $0.0015 per search, and says it is available immediately through its API, TypeScript SDK, command-line interface, and MCP server. That distribution matters for builders: a team can call it directly from an application, use it in TypeScript, invoke it from a shell workflow, or expose it to an MCP-compatible agent harness.

The launch follows Expand Fetch, a page-reading API that Expand says it released two weeks earlier. According to the announcement, Fetch uses models trained from scratch and preserves tables, icons, and hidden data. Expand positions Search and Fetch as adjacent parts of an agent research stack: Search finds relevant pages, while Fetch can supply more complete page content when an agent needs it. The company says Fetch is already handling millions of production requests, though the announcement does not independently document that volume.

Expand’s main sales argument is a vendor-run comparison of twelve configurations from eight providers, including Expand, Exa, Brave, Perplexity, Parallel, DeepSeek, Tavily, and Firecrawl. It combines five public benchmarks—WebCode, SimpleQA Verified, FRAMES, SEAL-Hard, and AA-Omniscience—into an equal-weighted composite index covering 2,985 questions. Expand says it ranked highest on that index and was Pareto-optimal on its tested price-versus-accuracy frontier. The company describes its score as comparable to Exa Auto and Brave LLM Context at less than one-third of their listed search cost.

The post also gives selected per-benchmark comparisons. On AA-Omniscience, a 600-question set, Expand reports 90.7% answer accuracy, $4.62 in search spend per 1,000 questions, and an average of 3.1 searches per question. It reports Exa Lite at 89.5% and $13.56, Brave at 89.8% and $19.68, Exa Auto at 89.7% and $22.68, and Exa Fast at 88.3% and $23.75. Against Perplexity Fast, Expand reports a higher score on SEAL-Hard and FRAMES for an extra $0.0005 per search. Against Firecrawl, it attributes part of its claimed advantage to richer excerpts: 523 tokens of evidence per result on average in its runs, compared with 112 for Firecrawl. All spend figures exclude model costs.

## What the archive adds

This is a new provider launch in an already active agent-search category, rather than a restatement of a prior Expand story. Earlier Field Notes documented [DeepSeek’s public Search and Fetch APIs](https://newruntime.com/posts/search-and-fetch-apis-now-public-at-0-5421c8/) and [Cloudflare’s web_search integration for the Agents SDK](https://newruntime.com/posts/new-web-search-tool-in-the-agents-sdk-9959ac/). Expand adds another API, with an MCP entry point and a published vendor comparison centered on search cost and answer accuracy.

## Trend signal

The launch offers limited support for [Agent economics moves to completed work](https://newruntime.com/patterns/agent-economics-moves-to-completed-work/). Expand frames price not only per call but per answered research question, accounting for repeated searches during an investigation. That is closer to task-level economics than token accounting, although the announcement still reports benchmark answer accuracy rather than verified production outcomes.

## Why now

Expand says users of its recently released Fetch API were frustrated with the search APIs they used alongside it. The timing therefore appears to be product-stack driven: after releasing a page-reading component, the company is offering a search component intended to sit before it in an agent’s retrieval flow.

## Connections to other events

The DeepSeek and Cloudflare notes show two different routes into the same builder problem: a search-and-fetch API pair, and search exposed inside an agent SDK. Expand combines another search API with a Fetch product and makes Search available through MCP, which lets an MCP-capable harness call it as a tool rather than requiring a bespoke integration.

## Why it matters

Search is a recurring operational cost for research agents because a task may require several queries, and a weak result may trigger follow-up search or page-fetch calls. Expand gives teams another service to evaluate at a transparent list price and through interfaces that fit common agent stacks. Its reported comparison also suggests a useful evaluation method: measure the cost of solving a representative question, not merely the price of one search. Include the number of search calls, the evidence returned in each result, subsequent fetches, model spending, answer quality, and human review time. A cheaper retrieval call can be a false saving if it forces more calls or produces evidence too thin for reliable downstream reasoning.

## Limits and uncertainty

The accuracy and cost comparisons come from Expand’s own testing and composite index, not an independent evaluation. The announcement does not provide enough information here to establish how the service performs on a particular company’s domains, languages, freshness requirements, rate limits, reliability needs, or safety constraints. The reported spend excludes model costs, so it is not a complete cost-of-answer measure. Benchmark accuracy also does not establish production usefulness for workflows that require citations, structured extraction, or human-approved decisions.

## Skill to build now

Create a small retrieval evaluation set from this week’s real agent tasks: a documentation lookup, a current-information question, and a multi-step research question. Run the same prompts through the current search provider and Expand Search. Log search count, returned evidence, required page fetches, model cost, final answer quality, and whether a reviewer accepts the output. This makes a provider choice an observable workflow decision rather than a response to a benchmark chart.

## Reader takeaway

Expand Search is a deployable new search option for agent workflows at $1.50 per 1,000 searches, with API, SDK, CLI, and MCP access. Its performance claims are promising but vendor-supplied. The useful decision is to test cost per accepted answer on your own tasks, including the retrieval and review work surrounding each search.

## Open question

Will Expand’s richer search excerpts reduce downstream fetches enough to preserve its claimed cost advantage in real agent workflows?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Mcp](https://newruntime.com/topics/mcp.md): Explore the mcp topic hub.
- topic hub: [Agent Workflows](https://newruntime.com/topics/agent-workflows.md): Explore the agent-workflows topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)