Energy, water & CO₂ footprint
Sourced, banded estimates of what your AI usage costs the planet, per model — for comparing choices, not for carbon accounting.
Claude Haiku 4.5
Anthropic
Energy / month
450 Wh
range 150 Wh – 1.3 kWh
~30 phone charges
low confidence · Third-party estimate (2025-10-01)
Water / month
525 mL
range 180 mL – 1.5 L
~3.5 s of a shower
low confidence · Third-party estimate (2025-10-01)
CO₂e / month
180 g
range 60.0 g – 600 g
~1.5 km driven in a petrol car
low confidence · Third-party estimate (2025-10-01)
Claude Opus 5
Anthropic
Energy / month
4.5 kWh
range 1.5 kWh – 12.0 kWh
~300 phone charges
low confidence · Third-party estimate (frontier-model scaling) (2025-08-01)
Water / month
5.0 L
range 1.7 L – 12.8 L
~4.95 L — about 0.6 min of a shower
low confidence · Third-party estimate (frontier-model scaling) (2025-08-01)
CO₂e / month
1.9 kg
range 675 g – 5.4 kg
~16.3 km driven in a petrol car
low confidence · Third-party estimate (frontier-model scaling) (2025-08-01)
These are rough, public estimates, not measurements. Published figures for AI energy, water and carbon vary by 10× or more depending on the model, hardware, data-centre efficiency and electricity grid, and only a few providers have disclosed measured numbers. Treat the range, not the middle value, as the answer, and see each model's page for sources and dates. Useful for comparing choices (for example routing easy requests to a smaller model) — not for carbon accounting or reporting.