Five vendors, five different layers: speed, index quality, distribution, trust, and time travel define agentic web search

Asked (summary):

Compare Exa, Parallel, Tavily, Nimble and Keenable using only information published in the last 90 days: major product announcements, AI features, partnerships, pricing or packaging changes, and customer or market signals. Summarize the biggest competitive difference among the established trio (Exa, Parallel, Tavily), name one sales risk and one sales opportunity per company, cite sources, and label inference.

Evidence covers 47 published items from 32 URLs across 21 hosts, all published 10 July – 8 October 2026, spanning all five companies and all five evidence categories. 30 of 47 items are company first-party claims; the rest come from independent news, analysis, or partners. Some items published in the window describe earlier events (noted in the timeline). All sales risks and opportunities below are labelled inference.

Competitive positioning matrix

Hover a column to focus one company. Amber tags mark inference; everything else is sourced fact. The matrix compresses each cell — the evidence table below carries full citations.

Evidence base per company: first-party vs independent

Count of supporting evidence items per company; green segment = items from independent hosts.
first-party / partner itemsindependent hosts backing the company

What the matrix cannot say alone

Tavily reported more than 2.5 million developers, up from 1 million in February — the strongest adoption signal in the window — alongside its first full quarter inside Nebius. agenticsearch.cloud
Parallel's Gemini Enterprise grounding integration bills via Google Cloud Marketplace or Bring Your Own Key with a shared default quota of 200 prompts per minute — procurement on the customer's existing Google invoice. docs.parallel.ai
Only two explicit packaging changes appear in the window: Tavily's keyless pay-per-search launch and Nimble's coming custom-budget "max" tier (currently gated); Exa's $7/1k Search and Keenable's 100,000-request free month are current prices, not changes. answers.databricks.com
Exa's benchmark leads over Tavily and Firecrawl (75.5% vs 40.5% rank-one people recall; 472 ms vs 1.4 s) are entirely company first-party, as are Nimble's 30–50% accuracy claims — buyers should weigh them accordingly. dev.to

The core three: where the value sits

Exa — retrieval quality and flexibility

Proprietary semantic index with composable endpoints (Search, Contents, Answer, Deep Search, Monitors) and entity-oriented strength; self-published benchmarks claim large rank-one recall leads at 472 ms median latency. Search listed at $7 per 1,000 queries.

Watch: all benchmark advantages are first-party; Search is rate-limited at 10 QPS.

Parallel — speed and enterprise cloud channel

Turbo completes retrieve-and-rank in a company-stated 200 ms (down from a 3-second budget), and the Gemini Enterprise integration puts Parallel on Google Cloud invoices. Newly published context: $100M Series A and $100M Series B (Sequoia) to a $2B valuation — financing events predating this window.

Watch: Turbo does not support domain/path source-policy filters.

Tavily — distribution and integrated AI cloud

More than 2.5M developers, keyless pay-per-search for autonomous agents, and full-stack integration with Nebius pairing Token Factory inference with real-time retrieval, plus NVIDIA usage in training and evaluation work.

Watch: answer endpoint adds LLM-token billing on top of $5–8/1k search.

Biggest difference (inference): the locus of value. Exa sells retrieval quality on its own index; Parallel sells latency plus the Google Cloud/Gemini channel; Tavily sells developer distribution and an integrated reasoning-plus-retrieval platform. The challengers attack different layers again: Nimble with per-claim trust grading and lakehouse-native integration, Keenable with Time Machine point-in-time retrieval. This synthesis is an inference from the cited facts.

Evidence timeline — 47 items, 10 July to 8 October 2026

company / partner first-partyindependent news or analysis — hover a dot for the fact and source

All 47 evidence items

PublishedCompanyCategoryFactRoleSource

Method: 47 supporting evidence rows for five agentic web-search vendors, drawn from 32 distinct URLs on 21 hosts, all published 2026-07-10 to 2026-10-08; the most-cited single URL backs 3 rows (6.4%). Counts are evidence items, prices are vendor-listed list prices per stated unit, and benchmark figures are reproduced as published (mostly first-party). Underlying event dates are shown where they differ from publication. Long facts are truncated in the matrix and table for space; sales risks and opportunities are analyst inference, not reported facts. keenable.ai

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.

Keenable SELECTAsk your own question
Made with Keenable SELECT