Wicklee / Inference Cost Lab

Wattage-per-Token
Calculator

The real cost of local AI inference — measured in watts, not just dollars. See exactly where local beats cloud and by how much.

GPU Preset
GPU TDP watts
350W
GPU Utilization %
85%
Tokens / Second tok/s
45tok/s
Electricity Rate $/kWh
$0.12/kWh
Watts / 1K Tokens
active power draw
Cost / 1M Tokens
electricity only
Daily Power (24h)
continuous inference
vs Cloud API — Cost per 1M Tokens
GPT-4o (OpenAI) $15.00
Claude Sonnet (Anthropic) $3.00
Gemini 1.5 Pro (Google) $3.50
Llama 3 via Groq $0.59