Search Provider Benchmark

Which search actually sees the whole market?

Six providers, eight scenarios. Beyond everyday research quality, we measure landscape recall — whether a search surfaces every real competitor in a field, including the obscure long-tail that naive agent fan-out routinely misses.

Recommended daily driver

Exa 86.69 Highest average score across all eight scenarios — and the widest lead on landscape discovery.

Overall ranking · average score

Exa
86.69
Perplexity
73.45
Tavily
60.41
Gemini (Google)
52.22
Grok / X
50.12
OFOffline fixture
26.05

Landscape discovery · entity recall

Each landscape is a curated roster of real products, tiered by obscurity. Long-tail entries are weighted highest — they are what a shallow search drops. Weighted recall is the headline number; the bar shows the best provider per market.

MCP server ecosystem

21 entities · fast-moving long tail

88%Exa · best recall
Exa
19/21 · lt 8/10
Perplexity
17/21 · lt 6/10
Tavily
15/21 · lt 4/10
Gemini
13/21 · lt 3/10
Grok / X
11/21 · lt 2/10
OFOffline
3/21 · lt 0/10
Missed by allJina AI MCP ToolsSerper MCP Server

AI accounting, close & audit automation

19 entities · SynthGL-adjacent market

86%Exa · best recall
Exa
17/19 · lt 7/9
Perplexity
15/19 · lt 5/9
Tavily
13/19 · lt 3/9
Gemini
11/19 · lt 1/9
Grok / X
8/19 · lt 0/9
OFOffline
3/19 · lt 0/9
Missed by allChatFinMaxima AI

AI spreadsheet copilots & agents

18 entities

86%Exa · best recall
Exa
16/18 · lt 7/9
Perplexity
13/18 · lt 4/9
Tavily
11/18 · lt 2/9
Gemini
10/18 · lt 1/9
Grok / X
8/18 · lt 2/9
OFOffline
3/18 · lt 0/9
Missed by allBardeenKuse

Excel I/O library landscape

17 entities · Wolf XL market

85%Exa · best recall
Exa
15/17 · lt 6/8
Perplexity
11/17 · lt 2/8
Tavily
10/17 · lt 2/8
Grok / X
7/17 · lt 2/8
Gemini
7/17 · lt 0/8
OFOffline
4/17 · lt 0/8
Missed by allClosedXMLEPPlus

AI financial diligence (QoE / FDD)

13 entities · SynthGL core market

79%Exa · best recall
Exa
11/13 · lt 4/6
Perplexity
8/13 · lt 2/6
Tavily
7/13 · lt 0/6
Gemini
5/13 · lt 0/6
Grok / X
4/13 · lt 0/6
OFOffline
3/13 · lt 0/6
Missed by allPeonyPinpointAI

Everyday research · best per scenario

Current events
Perplexity
Technical docs
Perplexity
Social sentiment
Grok / X

On focused research, the field is close — Perplexity leads news and docs, Grok wins live social. Landscape recall is where they separate.

How recall is scored

1

Each landscape is a verified roster of real products, each tagged headline / known / long-tail by how reliably a broad search surfaces it.

2

A provider's hits are scanned for every entity by name, alias, and domain. Weighted recall counts long-tail finds 3× a headline find — rewarding depth, not just naming the obvious.

3

Entities found by no provider surface as “missed by all” — systemic blind spots that tell you when the entire fan-out is failing a market segment.