Benchmarks
Does it actually work? We ran the numbers.
5 benchmark runs · 80+ prompts · real 92-file production codebase · same model (Claude Sonnet 4.6), same questions — with and without GrapeRoot.
Cost Evolution
Cost over version iterations
Category Cost
Cost by task category
Quality Scores
Quality across versions
Wall Time
Latency comparison
Token Volume
Input + output token usage
Cumulative Cost
Running cost across prompts
Win Rate
Win/loss/tie breakdown
Cost vs Quality
Every prompt plotted
Version Delta
Change vs prior version
Complex Prompts
Hardest prompt subset
Efficiency Radar
Multi-dimensional comparison
Challenge: Cost
Cost on challenge set
Challenge: Quality
Quality on challenge set
Challenge: Efficiency
Efficiency on challenge set
Challenge: Savings
Savings on challenge set
Full Cost Evolution
All versions, all costs
v3.8.39: 3-Way Cost
Three-way cost comparison
v3.8.39: 3-Way Quality
Three-way quality comparison
v3.8.39: Cost Improvement
Cost delta from prior version
v3.8.39: Win Rate
Win rate for v3.8.39
v3.8.39: Full Evolution
Full version cost evolution
v3.8.39: Before/After
Before/after comparison
