Tilman Bayer · Jan 20

I sure hope you are working from home, or if not, that your commute via electric car is less than 2.5 miles 😉

Dustin Moskovitz · Jan 21

I’d have to check the math on this but I’d guess nearly half the queries are below the median, even if yours happen to be above. So it balances out.

tj mahr 🤘 · Jan 20

so a bit more than running a kilolaptop for five minutes

Mike Janoski · Jan 21

Hi Simon, excuse the probably obvious question, but what sorts of benefits/capabilities do you get from your tooling?

Mike Wasson · Jan 20

So, depending on where you are, $0.20-$0.50 a day of energy usage. That doesn’t seem too bad for such a professionally useful tool.

Pilgrim · Jan 21

I think AI is a waste of natural resources. A complete waste. I know some people think it can help. I highly doubt it.

Lino Le Van · Jan 20

cc @andymasley.bsky.social @simonwillison.net might be interesting for you guys

Dr Vincent Raoult · Jan 20

They pick the median probably because it makes them look better. In this case it's probably better to use an average...

Josh You · Jan 20

hey, nice work! what does your (cached) input:output ratio look like? I don't strongly vouch for our estimates of input/attention costs still being relevant today so would be useful to know how sensitive this is to inputs vs outputs

Libbie Zorden · Jan 20

All right all right. I can see I'm going to have to educate myself on this s***, but I sure as hell don't want to.

Tyler Ham · Jan 21

I'd love to see an estimate of how many hours of work a coding agent saves, then it would be easy to calculate calories and then energy & water savings from a standard diet. Ofc, you're not actually saving energy but it'd be like taking a car a long distance vs walking the same place.

Musings, Sometimes · Jan 20

Great work - I skimmed through and will probably dive into the numbers deeply later. One also has to wonder if the Gas Town Crowd is looking at similar analyses!

Andreas Møller · Jan 21

I am pretty sure that the. OST of training is not included in the numbers Google /openai released for energy consumption per query so the real number is likely much bigger

Victor Geislinger · Jan 20

Thanks for sharing & it's super interesting to see from someone not at the 'median'! And not to discount what you found but it seems this might be more about the folks who use these tools frequently. And maybe that's worth exploring by itself since more people are using it like you (writing code)

Michael Martinez · Jan 20

I was told by @cjtrowbridge.com that this is not the case.

Simon P. Couch · Jan 20

cc @caseynewton.bsky.social and @kevinroose.com -- feels related to recent conversations on Claude Code + water/electricity use of LLMs

Alex Bradbury · Jan 20

Nice post! You might also be interested in my attempt to get some figures on inference energy usage, but coming from the perspective of the concrete data available in the InferenceMAX benchmarks. muxup.com/2026q1/per-q... Many many provisos and limitations of course

davepermen · Jan 21

So you're saying what you do is way more horrible than the median. And yet you still do it. Shameful.

Bartek Ogryczak · Jan 21

One thing I would add to this comparison would be the actual machine you use for coding, including the power consumed by external monitor.

Florent Bédécarrats · Jan 20

Thanks for the analysis! This does not include model training, right? Do we have a notion of how much the training represents in terms of energy consumption?