AI Efficiency: A Gemini Reality Check with Greenly
Summary
An analysis of Google's Gemini environmental report, critiquing its methodology and comparing results with Mistral AI and ChatGPT estimates.
Key quotes
Google reports emissions using a market-based approach... under a location-based approach... emissions would be closer to 0.09 gCOˢe per prompt - roughly three times higher.
without harmonized prompt specs and clear system boundaries, direct comparisons will remain misleading.
The article evaluates the environmental transparency of major AI models, noting that Google’s self-reported figures for Gemini often rely on market-based emissions rather than location-based ones. It argues for industry-wide standardization in measuring energy, water, and carbon footprints.