AgentLLM Cost & Routing Evaluator

v2.4 Pinned Benchmark
Model A Nominal
$126.00
Model A Effective
$226.80
Model B Baseline
$210.00
Optimized Hybrid
$142.50
Effective Savings
37.2%
Monthly Spend: Nominal vs Effective Output Inflation Agent LLM Evaluator — Live D3 Comparison
13-Benchmark Domain Performance Matrix (Opus 5 vs Fable 5 Baseline) 8 Wins / 5 Losses
Task Domain Task Share Opus 5 Win Rate Nominal Spend Effective Spend Recommended Model
Optimal LLM Router Map (Cost & Benchmark Maximized)

Assigned model routing based on benchmark win suitability vs verbosity inflation penalties:

JSON Deployment Config Preview:


          

        

      


    


  


  
  


  



Fixture prices and fixed routing determine a local cost comparison
Read the explanation
Ten thousand assigned monthly tasks use fifteen hundred input and eight hundred base output tokens each, totaling fifteen million input and eight million output. Verbosity one point eight raises output to fourteen point four million. At twenty pixels per million tokens the bars are three hundred, one hundred sixty and two hundred eighty eight. The named model prices are saved fixture constants, not verified current products or provider quotes. No actual model requests or bill measurements occur in these equations. Model A assigns input fifteen and output seventy five currency units per million tokens. Nominal cost is two hundred twenty five input plus six hundred output, totaling eight hundred twenty five. Effective verbosity raises output cost to one thousand eighty, giving one thousand three hundred five total. Model B assigns ten and fifty with verbosity one, totaling one hundred fifty plus four hundred, or five hundred fifty. At one pixel per five currency units the three total bars measure one hundred sixty five, two hundred sixty one and one hundred ten. The cost increase comes only from assigned output quantity and rates. Routing recommendations are fixed for five task domains. Browser and complex coding share fifty five percent and use model A, reasoning twenty percent model B, documents fifteen percent another fixture and chat ten percent a small fixture. The distribution sliders are independent and can sum above or below one hundred; hybrid cost does not normalize shares. At two pixels per percentage the three grouped bars fifty five, twenty and twenty five measure one hundred ten, forty and fifty. Header, table and routing JSON update before D3 chart rendering fails offline, so native inputs and CSV can still operate. Recommendations and benchmark win labels are assigned data, not validated capability comparisons or measured savings.


Super generates helpful tools and automates fact-checking across the internet proactively. If you enjoyed this tool, build your own with Super and share it with a friend.