Performance Benchmarks — Updated Daily

ConceptNet vs OpenAI & Claude

Real latency, real cost, real accuracy — on enterprise voice workflow tasks. No cherry-picking.

Last benchmark run: calculating...   live
5.2×
Cheaper than GPT-4o
2.1×
Faster intent classification
200
Languages supported
Intent Classification Latency (ms)
Task Type OpenAI GPT-4o Anthropic Claude ConceptNet
Cost per 1,000 Queries (USD)
Query Type OpenAI GPT-4o Anthropic Claude ConceptNet
Annual Cost Calculator
5,000/day
500 min/day
200 employees
OpenAI
$0
Anthropic
$0
ConceptNet
$0
$0
Annual savings vs OpenAI
Intent Classification Accuracy by Layer
Intent Layer OpenAI GPT-4o Anthropic Claude ConceptNet
Layer 1 — Basic
"Schedule a meeting for Tuesday"
 96%97% ✓ 97%
Layer 2 — Context-Aware
"Send report when deal closes"
 89%91% ✓ 94%
Layer 3 — Predictive
"Alert manager before contract expires"
 71%74% ✓ 88%
Layer 4 — Autonomous
"Always update CRM after every call"
 58%62% ✓ 91%
Multilingual Coverage
ProviderLanguagesEnterprise-grade accuracy (>90%)Cross-cultural voice
OpenAI GPT-4o
~50~12❌ No
Anthropic Claude
~30~8❌ No
ConceptNet
✓ 200✓ 47✓ Yes (SONAR)
Methodology
Test corpus: 2,400 enterprise voice commands across 4 intent layers, sourced from publicly available customer service and enterprise workflow datasets. No proprietary data used.

Latency: Measured as time from input submission to structured JSON output, averaged over 100 runs per query type, excluding network overhead. ConceptNet runs locally; OpenAI/Claude times are API round-trip from London, UK.

Cost: Based on published API pricing as of July 2026. OpenAI GPT-4o at $0.005/1K tokens input, $0.015/1K output. Claude Sonnet at $0.003/$0.015. ConceptNet token-free architecture costs measured against equivalent cloud compute.

Accuracy: Human-labelled ground truth on intent layer classification. Single-label classification task per command.

Reproducibility: All test scripts and datasets available at github.com/wushu75/ConceptNet/benchmarks.