Day-one estimate
Calculated from an estimated parameter count — the maker hasn't published one. Every model on this site is calculated the
same way, so the comparison holds even where the absolute value carries uncertainty.
Our figures are good to roughly a factor of two to three, which is why we only ever
state a difference of 5× or more.
The figure above is an understatement. Claude Sonnet 4.6 generates reasoning text you never see, and that text costs energy to produce. Our calculation counts a fixed answer length, so the real figure is higher — we cannot say by how much, because the amount is not published.
Alibaba · capability 84.8. It scores in the same tier as Claude Sonnet 4.6 — 78–88, the same standard of work — and uses about 22× less energy per answer: 0.351 Wh against 7.65 Wh.
What Anthropic publishes
Energy cannot be worked out without knowing how big a model is. Of the 21 models
from Anthropic that we track, 10 can be assessed at all — the rest publish
too little for anyone outside the company to calculate anything.