AI Energy Calculator
How much electricity does your AI usage cost the planet? Pick a model, enter your usage, and see the numbers.
Per Query
6.00 Wh
Per Month
18.0 kWh
CO2 / Month
8.8 kg
kWh / Day
0.600 kWh
That's equivalent to...
5.9 min
of passenger flight time per month
42.0 mi
of driving an average car per month
1800 hrs
of a 10W LED bulb per month
Energy per query (all models)
| Model | kWh/Query | vs Google Search |
|---|---|---|
| GPT-4 | 0.0100 | 33x |
| GPT-4o | 0.0060 | 20x |
| GPT-4o Mini | 0.0015 | 5x |
| GPT-3.5 Turbo | 0.0020 | 7x |
| o1 | 0.0300 | 100x |
| Claude Opus 4 | 0.0110 | 37x |
| Claude Sonnet 4 | 0.0050 | 17x |
| Claude 3.5 Haiku | 0.0012 | 4x |
| Gemini 2.0 Pro | 0.0055 | 18x |
| Gemini 2.0 Flash | 0.0010 | 3x |
| Llama 3.1 405B | 0.0150 | 50x |
| Llama 3.1 70B | 0.0040 | 13x |
| Llama 3.1 8B | 0.0008 | 3x |
| Mistral Large 2 | 0.0060 | 20x |
| DeepSeek V3 | 0.0035 | 12x |
| DALL-E 3 | 0.0500 | 167x |
| Midjourney v6 | 0.0450 | 150x |
| Stable Diffusion XL | 0.0200 | 67x |
| Google Search (baseline) | 0.0003 | 1x |
Where these numbers come from
Exact energy per query is hard to pin down because inference costs depend on hardware, batch size, datacenter efficiency (PUE), and model optimizations. The numbers here are estimates based on the best publicly available data.
Key sources:
- IEA "Electricity 2024" report on AI energy consumption
- De Vries, A. (2023). "The growing energy footprint of artificial intelligence." Joule.
- Luccioni, A. et al. (2023). "Power Hungry Processing." NeurIPS.
- SemiAnalysis inference cost estimates
- Google Environmental Reports (2023, 2024)
- Epoch AI compute analysis
CO2 calculations use the global average grid carbon intensity of 0.49 kg CO2/kWh (IEA 2023). Your actual carbon footprint depends on where the datacenter is located. A query processed on France's nuclear grid (0.05 kg/kWh) produces 10x less CO2 than the same query on a coal-heavy grid (0.9 kg/kWh).
These numbers will change as hardware gets more efficient and models get optimized. I update them quarterly.