info
DEMO DATA: scores shown are template sample values pending independent evaluation · as of 2026-09-05 · Methodology
INDEX
/
AI_RESEARCH
/
PERPLEXITY
v2026.2-PROD
PROVED: 0x81b4..e99f
dns
US-EAST-EVAL-04
|
|
LATENCY: 120ms P95
Real-Time Web Retrieval
Multi-Model Router
Citation-Backed
Deep Research Engine
Perplexity
/ PRO SEARCH & DEEP RESEARCH
business
Perplexity AI (Aravind Srinivas / San Francisco, CA)
•
verified
Verified Deterministic Index
Conversational search index engineered with synthetic query tree decomposition, real-time index scrapers, and dynamic model orchestration across Claude 3.7 Sonnet, OpenAI o3-mini, and DeepSeek-R1 for fully-cited research dossiers.
Try Perplexity Pro
open_in_new
$
curl api.perplexity.ai/chat/completions
compare_arrows
Compare Matrix
Composite Index Rating
CLASS: S-TIER
95.2/100
Rank #1 in Research & Retrieval
trending_up
+18.2%
30d Velocity
Benchmark Confidence:
99.8% Deterministic
Sample Traces (30d):
1,420,891 requests
Primary Engine:
Sonar Deep-32k + o3
Citation Precision
link
98.4%
Multi-source cross-verification & semantic grounding
check_circle
99.2% alive target URIs
Deep Research Exhaustion
hub
94.7%
Autonomous query decomposition across 40+ sources
Avg Depth: 6 iterative passes
P95 Synthesis TTFT
speed
120ms
Groq LPUs & TensorRT-LLM server acceleration
bolt
164 tokens/sec stream
Hallucination Guard
format_image_left
99.1%
Zero phantom URL citations via Perplexity Web Index
verified
Strict RAG constraint
Empirical Telemetry Vectors
Standardized automated synthetic audits executed continuously over 30 days.
menu_book
Academic & Web Synthesis Depth
96.2%
Cross-domain ArXiv & Semantic Scholar indexation
+2.4% vs prev version
shuffle
Multi-LLM Routing Agility (Claude 3.7, o3, DeepSeek-R1)
97.5%
Seamless model swap latency < 25ms
Optimal task heuristic
fact_check
Source Attributability & Footnote Integrity
98.8%
Direct verbatim match against cited raw crawl payloads
Zero dead anchors
attachment
Multi-modal Ingestion (PDF, Charts, Tabular Data)
92.4%
OCR extraction fidelity for tabular financial filings
Supports 100MB / file
psychology
Agentic Deep Research Workflow Execution
95.1%
Multi-branch sub-question recursion and self-critique
5k+ word dossiers
30-Day Evaluation Latency & Citation Drift Index
Precision
Baseline
Day -30 (v2.1 Release)
Day -15 (Deep Research Deployed)
Today (98.4% Precision Peak)
Engine Architecture
Technical primitives powering the Perplexity inference pipeline.
01
Autonomous Deep Research Tree
Decomposes complex multi-faceted investigative prompts into autonomous exploration trees, analyzing dozens of web enclaves before synthesizing 5,000+ word structured dossiers.
Heuristic: Dynamic BFS Branching
02
Dynamic Model Fabric
Native in-flight model switcher including Claude 3.7 Sonnet, OpenAI o3-mini / GPT-4o, DeepSeek R1, and proprietary Sonar 32k/128k LLMs based on query complexity and mathematical depth.
Switch Latency: < 20ms in-pipeline
03
Live Web Graph & PerplexityBot
In-house fresh web crawler with millisecond URL indexing, bypassing stale static training data boundaries with real-time Perplexity Search API and index verification.
Recrawl SLA: 15-minute global delta
04
Spaces & Enterprise Enclaves
Granular document uploads (PDF, DOCX, CSV), collaborative workspaces, and zero-data retention policies (ZDR) preventing telemetry leakage into generic training corpora.
Compliance: SOC2 Type II, HIPAA, GDPR
tune
Head-to-Head Telemetry
Direct Peer Evaluation Matrix
Updated: Daily at 00:00 UTC
| Engine / Platform | Airecmark Score | Retrieval Paradigm | Deep Research Mode | Model Agility | Hallucination Delta | Pricing / Seat |
|---|---|---|---|---|---|---|
|
P
Perplexity
Target
|
● 95.4 | Proprietary Live Web Crawler + Multi-Index | check_circle Included (Autonomous 40+ src) | Claude 3.7, o3, DeepSeek R1 | -78.2% vs raw LLM | $20 / mo |
|
C
ChatGPT (Search)
|
93.8 | Bing Index + In-flight Scraper | check_circle Via o3-mini Deep Research | OpenAI Only (GPT-4o, o-series) | -69.4% vs raw LLM | $20 / mo |
|
G
Google Gemini Adv.
|
91.2 | Google Search Engine Native Graph | remove Single-query Grounding Only | Gemini 1.5 Pro / 2.0 Flash | -62.1% vs raw LLM | $20 / mo |
|
C
Consensus
|
92.0 | Semantic Scholar & PubMed Academic | check_circle Paper Extraction Synthesizer | Custom Domain RAG | -74.0% (Academic only) | $11.99 / mo |
|
N
Google NotebookLM
|
89.6 | Strict User Corpus Grounding (No live web) | remove Audio Overview Synthesizer | Gemini 1.5 Pro | -81.0% (Within corpus) | Free ($0) |
Metrics extracted via Airecmark Automated Crawler Harness #729.
Standardized across 10,000 multi-step enterprise queries.
Financial & Token Econometrics
Subscription & API Tiers
Verified plan specifications and monthly throughput allowances.
Standard Consumer
Free Tier
$0
/ forever
- check Unlimited Standard Quick Searches
- check 5 Pro Searches per 4-hour interval
- check Default Sonar Base LLM
- close Deep Research Agent Disabled
- close Claude 3.7 / o3-mini routing blocked
Best Value for Technical Teams
Engineering Tier
Perplexity Pro
$20
/ month ($200 billed yearly)
- check 600+ Pro Searches / day
- check Unrestricted Deep Research Engine Access
- check Claude 3.7 Sonnet, o3-mini, DeepSeek R1
- check $5.00/mo included API credits
- check Unlimited file & PDF multi-modal uploads
Sovereign Deployment
Enterprise Pro
$40
/ seat / month
- check All Pro Capabilities Included
- check Strict Zero Data Retention (ZDR) Enclave
- check SSO, Okta, SAML 2.0 & SCIM Provisioning
- check Shared Team Spaces with ACL management
- check Dedicated Enterprise SLAs & Support Lead
verified_user
Airecmark Lead Analyst Assessment
"Perplexity represents the single most successful paradigm shift in real-time information synthesis since modern search indexing. By grounding multi-tier generative models in deterministic URL crawls with strict attribution footnotes, it has effectively solved the hallucinated search problem for engineering and analytical research workflows."
EVALUATION SIGN-OFF:
SIGNER: 0x992B..F10A (AIRECMARK-DAO)
check_circle
Empirical Advantages
- › Dynamic switching between leading foundational models (Claude 3.7 / o3-mini / DeepSeek R1) without losing chat memory.
- › Deep Research automatically executes 40+ iterative web crawls, outputting publication-grade comprehensive dossiers.
- › Extremely rapid time-to-first-token (TTFT) via Groq accelerated Sonar infrastructure.
warning
Documented Technical Trade-offs
- › Occasional paywalled academic publications fail automated crawler retrieval, falling back to static abstracts.
- › Search API token usage pricing can scale quickly during heavy agentic recursive batch runs.
lock_reset
Deterministic Evaluation Ledger Hash
sha256: 4f1a238e8984920b72c91834720938f903824abec91823749283749281c01b22