Significant change in impact of OpenAI models ? #199
-
|
Thank you for all the work on this great tool ! When using Ecologits for a study in September 2025, the order of magnitude was around 54 mgCO2e per token for gpt 4o (OpenAI). Today, the same model accounts for around 0.77 mgCO2e per token. Was there an update in the data which could explain this change ? For example for a "Small conversation (400 tokens)" :
This might be due to an error on my side back then, but the values are very different (does not look like an conversion error for example). |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
Hey :) This is due to a major update in our methodology released in around November 2025, see this blog post. We now account for more recent GPUs, batching and we consider a more efficient backend (vLLM), leading to lower approximations, but that we believe closer to reality. |
Beta Was this translation helpful? Give feedback.
-
|
Hi, thank you for the quick reply ! That's exactly what I was looking for, I missed it somehow. That's a major update indeed ! |
Beta Was this translation helpful? Give feedback.
Hey :) This is due to a major update in our methodology released in around November 2025, see this blog post. We now account for more recent GPUs, batching and we consider a more efficient backend (vLLM), leading to lower approximations, but that we believe closer to reality.