info
DEMO DATA: scores shown are template sample values pending independent evaluation · as of 2026-09-05 · Methodology
INDEX
/
CODING_AGENTS
/
WINDSURF
v1.4.2 PROD
PROVED: 0x7E19...B4A9
|
CLUSTER: US-EAST-EVAL-09
surfing
Windsurf
Codeium Inc. Tier 1 Sovereign IDENext-generation agentic developer environment driven by the Cascade dual-flow engine, real-time context graphs, and unified codebase synchronization.
CASCADE ENGINE
SUPERCOMPLETE
MCP INTEROP
SOC 2 TYPE II
CLAUDE 3.7 NATIVE
Composite Index
91.8
/100
trending_up
+14.2% 7d velocity
Global Category Rank
#3
of 84 Coding Agents
Top 3% Velocity Sector
SWE-bench Lite Verification
89.2%
pass@1 AST
1,240 synthetic tasks
P95 Inference TTFT
85ms
edge cache
2.4x faster than Copilot
24h Telemetry Runs
18,400
ops
99.98% runtime fidelity
Execution Pipeline
Cascade Dual-Agent Loop Engine
Unlike static copilots that invoke isolated generation prompts, Windsurf executes an uninterrupted synchronous loop between localized speculative tokens (Supercomplete) and deep contextual agent commands (Cascade).
AST PARSER → CONTEXT GRAPH
SPECULATIVE COMPILATION
Tier 01
check_circle
Local AST Graph
Semantic multi-file indexing sub-30s cold cache.
128k context slice
Tier 02
bolt
Supercomplete
Speculative intent prediction ahead of user cursor position.
85ms P95 latency
Tier 03
terminal
Cascade Terminal
Deterministic terminal execution and iterative patch self-healing.
Sandboxed runs
Code State
→
Cascade Multi-Diff
→
Patch Verification
Baseline RAM
~180MB
-31% vs Cursor
Cold Cache Index
24.2s
500k LOC repo
Context Cohesion
94.8%
Proprietary AST
MCP Tools
Native v1.0
Zero-config
Empirical Vector Score
92.6 OVERALL
Context Retrieval Precision (128k)
94.8%
Autocomplete Latency (P95 TTFT)
96.0%
Multi-File Agent Orchestration
91.4%
SWE-bench Lite Solution Rate
89.2%
MCP & Protocol Ecosystem Interop
92.0%
Integrated Model Roster
5 ACTIVE PROVIDERS
Claude 3.7 Sonnet
DEFAULT
Claude 3.5 Sonnet
FAST
OpenAI o3-mini
REASONING
GPT-4o Omnimodal
VISION
Codeium Proprietary Autocomplete Cluster
85ms SLA
Empirical Comparative Index
Coding Agent Tier-1 Matrix
SAMPLE RUNS: 1,000 PER AGENT | SEED: 0x9AF8
| Agent Candidate | Composite Score | P95 TTFT Latency | SWE-bench Lite | Base Pricing | Memory Footprint | Status |
|---|---|---|---|---|---|---|
| terminal Claude Code CLI Engine | 95.8 | 420ms | 92.4% | BYOK Token | <45MB | RESEARCH |
| code Cursor v0.45 | 94.2 | 110ms | 90.8% | $20/mo | 260MB | LEADER |
| surfing Windsurf (Codeium) TARGET | 92.6 | 85ms | 89.2% | $15/mo | 180MB | EVALUATED |
| memory Aider Git-paired | 92.4 | 380ms | 88.7% | Open Source | 60MB | OPEN SOURCE |
| smart_toy GitHub Copilot Workspace MSFT | 91.0 | 205ms | 84.1% | $10-19/mo | Standard | INCUMBENT |
Capital Allocation
Commercial Tiers & Token Economies
Starter
COMMUNITY
$0
/ forever
Unlimited fast speculative tab completions powered by Codeium's proprietary cluster.
check
Unlimited Supercomplete tab completions
check
25 Cascade deep agent flows / mo
check
VS Code extension parity
BEST VALUE ($5 DISCOUNT VS CURSOR)
Pro Developer
$15
/ month billed monthly
Full unlimited access to Cascade agent flows, Claude 3.7 Sonnet, and low-latency fast queues.
check_circle
Unlimited Cascade Agentic Flows
check_circle
500 Fast Frontier queries (Claude 3.7 Sonnet)
check
Zero model rate limiting fallback
check
MCP Protocol local server support
Enterprise
SOVEREIGN
$30
/ seat / mo
Zero Data Retention, Dedicated VPC inference clusters, SAML SSO, and audit compliance logging.
check
Zero Data Retention SLA (SOC 2 Type II)
check
Custom VPC endpoint deployment
check
Centralized admin seat management
verified
Airecmark Analyst Consensus
“Windsurf’s Cascade architecture bridges the gap between passive autocomplete and full-blown agentic delegation, offering arguably the smoothest user flow and highest value-for-money at $15/month.”
thumb_up
Empirical Advantages
- • Cascade UX avoids clunky multi-window context switching with real-time in-line execution loops.
- • Aggressive pricing strategy ($15/mo vs Cursor's $20/mo) without compromising on Claude 3.7 Sonnet access.
- • Measurably lighter RAM footprint (~180MB baseline vs Cursor's 260MB).
info
Documented Trade-offs
- • Third-party VS Code extension marketplace integration occasionally trails upstream VS Code updates by 2-3 weeks.
- • Extremely complex multi-repo refactoring falls slightly behind Claude Code terminal CLI in pure AST consistency.
SOURCE: AIRECMARK CLUSTER NODE #09 • SHA256: 8C14F2...E89D
LAST BENCHMARK SWEEP: TODAY 04:12 UTC • RUNTIMES: PYTHON, TYPESCRIPT, RUST, GO