PAIR INDEX v2.4

AiRecMark / Compare / Pairwise Benchmark Terminal

SOURCE: PUBLISHED ARCHIVES SEED: data/tools/*.json

Head-to-Head Comparison

SNAPSHOT 2026-09-06

Pairwise comparison computed from the published five-dimension rubric (quality 25% · features 20% · usability 15% · performance 20% · value 20%).

terminal

Target Asset A INCUMBENT

Cursor

AnySphere Inc. • Electron Shell

Swap unfold_more

DIFF GAP

code_blocks

Target Asset B CHALLENGER

Windsurf

Codeium

Swap unfold_more

Benchmark Presets: Claude vs ChatGPT Lovable vs Bolt.new Claude Code vs Cursor Perplexity vs Gemini

verified Consensus Terminal Memo & Alpha Differential

SOURCE: PUBLISHED ARCHIVES RUBRIC: v2-5dim DIMENSIONS: 5

Cursor Institutional Score HIGHER RUBRIC SCORE

89.2

Spread Efficiency Index RATIO 1.063

Share of combined score Cursor 89.2 vs Windsurf 83.9

Composites of the published five-dimension rubric

Windsurf Institutional Score BASELINE

83.9 Baseline

analytics

Analyst Memorandum Verdict: Retain Overweight Cursor for Monorepo Systems

Cursor secures statistical alpha (+2.4 pts) in large workspace AST indexing velocity and Model Context Protocol (MCP) tooling flexibility. Windsurf captures parity on localized flow speed and agentic multi-file Cascade execution, but exhibits context degradation beyond 64k token windows.

Pairwise Differential Archive record Matrix

Zero-centered bidirectional differential tracks. Centerline marks absolute parity (0.0).

Cursor Lead

Windsurf Lead

1. Context Recall @ 64k Tokens

NEEDLE-IN-HAYSTACK 64-PASS

96.4%

82.1%

Dynamic RAG window cache Proprietary buffer compaction

2. Cascade Flow & Multi-file Edit Execution WINDSURF +1.8%

PARALLEL AST REWRITE

88.0%

89.8%

Speculative diff generator Cascade synchronous AST engine

3. MCP Standard Protocol Compliance

ANTHROPIC MCP SPEC V1.0

100%

28.0%

Full custom client & server bridges Partial tool mapping only

4. Time to First Token (TTFT Latency)

LOWER IS BETTER • P95 MEDIAN

420ms

450ms

US-East Proxy cluster Direct Exafunction routing

5. Monorepo Indexing Velocity (500k+ LOC)

EMBEDDING THROUGHPUT

1.8x

1.0x (Base)

Rust worker multi-threading Node worker pools

6. Enterprise 90-Day Retention Rate

ANALYST SURVEY N=3,200

89.4%

84.2%

Cohort churn: 1.2% mo/mo Cohort churn: 2.1% mo/mo

archive-recorded Specification & Governance Differentials

Empirical runtime constraints, licensing boundaries, and pricing parity benchmarks.

Parameter Matrix	Cursor (AnySphere)	Windsurf (Codeium)
Core Runtime Architecture	Custom VS Code Fork (1.93.1 base) with C++ Semantic Indexer	Custom VS Code Fork with Cascade Proprietary Engine
Pro Pricing Tier	$20 / user / month	$15 / user / month (-25% delta)
Fast Request Allowances	500 Fast Requests/mo (overages at standard queue)	Unlimited Flow actions with dynamic rate-limiting
Supported Model Roster	Claude 3.7 Sonnet, Claude 3.5 Haiku, GPT-4o, o3-mini, DeepSeek R1, Custom OpenAI Base URL	Claude 3.7 Sonnet, GPT-4o, Codeium Supercomplete Base, Custom API Keys
Zero Data Retention (ZDR)	check_circle SOC 2 Type II + Privacy Mode Enabled	check_circle SOC 2 Type II + Enterprise Zero Storage
MCP Server Interop	Full Support (Cursor Settings > Features > MCP)	Experimental / Internal Cascade Only

LAST ARCHIVE SNAPSHOT: 12 MIN AGO

Related Contenders in Coding Agent Infrastructure

High-signal substitutes benchmarked under the same institutional protocol.

View All 42 Code AI Contenders arrow_forward

Terminal CLI Agent Score 86.1

Claude Code

Anthropic's native command-line agent with deep direct bash integration and sub-agent task orchestration.

Differential vs Cursor

Load into Terminal

Enterprise Standard Score 88.1

GitHub Copilot Workspace

Microsoft-backed institutional deployment standard with zero-liability indemnification clauses.

Differential vs Cursor

Load into Terminal

Cloud Containerized Score 81.8

Replit Agent

Zero-setup cloud runtime execution pairing full-stack scaffolding with integrated PostgreSQL deployment.

Differential vs Cursor

Load into Terminal