info
DEMO DATA: scores shown are template sample values pending independent evaluation · as of 2026-09-05 · Methodology
RANK #1 IN CODING AGENTS
Cursor
verified v0.45.2 Tier 1 SovereignDeterministic, multi-file agentic IDE engineered for monorepos, AST code intelligence, and headless model orchestration.
VENDOR: Anysphere Inc.
•
LICENSE: Proprietary / VS Code Fork
•
SECURITY: SOC 2 Type II Verified
•
LAST BENCHMARK: Today, 04:18 UTC
Empirical Performance Audit
Macro Intelligence Index
94.2
/100
Alpha Winner
Relative Velocity
Top 1% in Sector
↑ 3.2σ vs Baseline
Telemetry Runs
14,280 daily runs
Cluster: US-East-1
Synthetic Hallucination
0.00% TS/C++
N=4,500 AST tests
Coding AST Correctness & Spec Execution
(SWE-Bench Lite / Multi-repo)
97.2%
Tier 1 Deterministic
P99 Pass-Rate: 88.4%
Long Context Recall & Monorepo Indexing @ 128k
(Mermaid Graph Walk)
96.4%
Local C++ SQLite Embeddings
Lossless @ 200k Window
Execution Speed & Latency (Time to First Token)
(P95 TTFT)
92.1%
P50: 160ms • P95: 240ms
Streaming Speculative Decoding
MCP (Model Context Protocol) & Ecosystem Interop
(Client + Server Bridge)
100.0%
Full Anthropic MCP Specification
Native JSON-RPC Bridge
Enterprise Governance, Zero Data Retention (ZDR)
(Privacy Compliance)
85.0%
ZDR available via Business plan
Requires custom DPA
Low-Level Verification
ISO/IEC 25010 Ref
Quantitative Technical Specification
terminal
Runtime & Core Shell
Custom VS Code Upstream Fork
Built on Electron with deeply embedded native C++ semantic background threads and Rust AST parser to bypass standard JavaScript IPC bottlenecks.
Binary footprint: ~210MB
hub
Context Indexing Engine
Dynamic Multi-File RAG + Merkle Hash Sync
Calculates AST graph diffs on local saves. Employs 768-dim embeddings with 200,000 token active memory slice without remote repo leaking.
Index latency: <45s for 500k LOC
device_hub
Protocol Specification
Full MCP Client / Server Protocol Bridge
Allows bidirectional execution across local databases (PostgreSQL/SQLite), shell tools, and remote server endpoints compliant with Anthropic standards.
Spec Compliance: MCP v1.0.3 Verified
smart_toy
Model Agility & Routing
Multi-LLM Sovereign Execution
Instant toggling between Claude 3.7 Sonnet (Hybrid Thinking mode), GPT-4o, OpenAI o3-mini, DeepSeek R1, and custom proxy endpoints via Base URL.
BYOK (Bring Your Own Key): Supported
Market Positioning Matrix
Open Full Matrix open_in_new
Direct Competitor Radar & Alternatives
| Tool & Version | Score | Primary Sweet Spot | Entry Pricing | Action |
|---|---|---|---|---|
|
C
Cursor
(Current)
|
94.2 | Monorepos & deep multi-file diff generation | $20/user | Selected |
|
W
Windsurf v1.1
|
91.8 | Agentic Cascade & automated terminal fixes | $15/user | Compare vs |
|
CC
Claude Code CLI
|
93.8 | Pure terminal headlessness & batch refactor | API Direct | Compare vs |
|
GH
GitHub Copilot
|
91.0 | Enterprise FedRAMP & GitHub ecosystem | $19/seat | Compare vs |
|
R
Replit Agent
|
88.4 | Autonomous full-stack scaffolding in browser | $25/user | Compare vs |
Institutional Synthesis
Analyst Consensus Verdict
“Cursor v0.45 establishes an unmatched deterministic benchmark for multi-file repo refactoring. By moving context indexing into a dedicated native process rather than a basic language extension, Anysphere captures zero-latency AST graph navigations that competitors struggle to replicate. While subscription lock-in for enterprise policy remains a trade-off, its performance delta justifies engineering adoption.”
— Airecmark Senior Systems Architecture Committee
verified
Verified Strategic Strengths
- ✓ Native Composer Interface: Allows orchestrating simultaneous multi-file edits with instant inline diff previews.
- ✓ Granular Model Switch: Zero switching penalty between Sonnet 3.7, o3-mini, and local Ollama weights.
- ✓ MCP Tool Ecosystem: Seamless integration of terminal executions and external API endpoints directly in context.
- ✓ Semantic Codebase Indexing: Lossless local vectors ensure sensitive IP stays local on Developer tier.
warning
Institutional Trade-offs & Headwinds
- ! Fork Lag Risk: As a standalone VS Code fork, updates can occasionally trail official Microsoft releases by 2-4 weeks.
- ! Enterprise Governance Paywall: Zero Data Retention (ZDR) SLA and audit logs require the $40/seat tier.
- ! Token Rate-Limits: Premium "Fast Requests" capped at 500/mo on Pro, after which queue latency increases during peak hours.
- ! Memory Footprint: Indexing repositories >2M LOC can push local RAM usage above 3.5GB.
Total Cost of Ownership
Transparent
Commercial Tiers
Hobbyist
$0 / mo
2,000 basic AI completions, 50 slow premium requests.
Data policy: Standard telemetry
Most Deployed
Developer Pro
$20 / mo
500 fast premium requests (Sonnet 3.7/GPT-4o), unlimited slow requests, 10 o3-mini fast runs/day.
Privacy: Opt-out data storage included
Business / Org
$40 / user
Centralized seat management, workspace privacy mode, zero retention SLA, SAML SSO, admin telemetry.
SOC 2 Type II + DPA Guaranteed
TCO MODEL NOTE: Overrun usage beyond 500 fast requests is billed at exact API provider pass-through rates ($0.003/1k input tokens on Claude 3.7 Sonnet). Median enterprise engineer consumption runs ~$28.40/mo.
Audit Provenance
Evaluation Metadata
AUDIT HASH
0x9f4a...e12d
TEST REPO LOC
1.42M Lines (Rust/TS)
HARDWARE ENVIRONMENT
Apple M3 Max (64GB)
OS RUNTIME
macOS 15.2 / Ubuntu 24.04
AIRECMARK AUDITOR
Cluster node-us-east-4
Developer Sentiment
Ecosystem Gravity
Developer NPS Score
78/100 (Exceptional)
Daily Active Share in Agentic IDEs
64.2% Dominance
Sources: GitHub, Discord, SWE-Pulse
N=18,900
shield_with_heart
AIRECMARK EMPIRICAL PROTOCOL v2.4
•
Independent, un-sponsored deterministic evaluation clusters.
All benchmarks executed in sandboxed hypervisors with standardized token latency meters.