Skip to content

Benchmarks Results

Devrajsinh Gohil edited this page Aug 30, 2026 · 1 revision

Master Benchmark Results Matrix

Validated across three independent experimental tiers on Apple Silicon (M-Series ARM64, 8 Cores, 16GB RAM):

Benchmark Layer Topology Profile Sample Size LangGraph Native (Mean) AgentMesh C++ (Mean) Paired Advantage (Δ) Win Rate Statistical Significance
Tier 1: Synthetic 1,000 In-Memory Hops N = 100 3.80 ms / hop 0.002 ms / hop +3.80 ms (1,900x) 100% p < 0.0001 (Deterministic)
Tier 1: Synthetic Diamond DAG Barrier N = 100 15.60 ms 0.012 ms +15.59 ms (1,300x) 100% p < 0.0001 (Deterministic)
Tier 2: Replay 8-Way Wide Fanout Swarm N = 50 197.88 ms 187.97 ms +9.91 ms (+5.0%) 90% (45/50) t = 9.78, p < 0.001
Tier 2: Replay Sequential Reflection Swarm N = 50 284.69 ms 277.00 ms +7.69 ms (+2.7%) 78% (39/50) t = 6.61, p < 0.001
Tier 3: Live Cloud Groq LPU Swarm ($AAPL) N = 5 7,065.33 ms 6,807.29 ms +258.04 ms (+3.7%) 80% (4/5) Live WAN Dominated
Tier 3: Tail (P95) Groq LPU Tail Latency N = 5 8,518.31 ms 7,628.35 ms +889.96 ms (+10.4%) 80% (4/5) P95 Stability Win
Memory Density 100 Concurrent Workflows N = 100 240.0 MB 12.0 MB -228.0 MB (95% Less) 100% Constant C++ Stack Alloc
Crash Recovery Interrupted Task Resumption N = 20 Crash / Data Loss 4.8 ms Instant Resumption 100% PostgreSQL libpq WAL

Clone this wiki locally