Anthropic

How much water does Claude use?

Anthropic has published no per-query water or energy figures. Here is what can be estimated from independent benchmarks, scenario by scenario.

None

official per-query disclosure published

~0.8 Wh

per 1K tokens, Claude Sonnet class (benchmark mid)

344 L

for 86M tokens of Claude Sonnet at Mid, about 5 showers

Anthropic has not published per-query energy or water figures for Claude, and its infrastructure runs on cloud providers (AWS and Google), whose water shows up in those companies' ledgers. So every Claude number on this site is an estimate from independent benchmarks, and is labeled as such.

By model tier, we class Claude Haiku as efficient, Claude Sonnet as standard, and Claude Opus as frontier, the same banding Jegham et al. use for comparable models. At Mid, a Sonnet exchange of about 700 tokens works out to roughly 3 mL of water; Opus several times that; extended thinking modes more still.

Claude is also where the developer story is most concrete: coding agents report token usage directly (run /cost in Claude Code). A heavy multi-month agent run of 86 million tokens computes to 344 liters at Mid: about five showers. That number is the reason this calculator has an 'I know my tokens' tab.

If Anthropic publishes an official figure, it gets an override in our coefficients file the same week, like Google's did.

Your number

What does your AI actually use?

Pick your model, set your usage, get the number, with sources.

Try the calculator

Sources

Cited on this page, in order of appearance.