https://olafnub.substack.com/p/how-much-energy-is-consumed-in-different
Not all models are listed (least → greatest):
| Tool | Model Size | Energy per Query | Notes |
|---|---|---|---|
| Google AI (without AI suggestion) | Small models | ~0.3 - 1 Wh | Lightweight, optimized for speed & low power |
| Perplexity.ai | Mix of models | ~2 - 8 Wh | Uses GPT-4, Claude, custom models |
| Claude (Anthropic) | ~Claude 3 family | ~3 - 8 Wh | Generally tuned for efficiency vs GPT-4 |
| ChatGPT (GPT-4o) | ~1.8T params* | ~5 - 10 Wh | Depends on query length/context |
TLDR:
Limit word count: Ask for "<X words" or "1-2 sentences."
Example Prompts:
Avoid background context (unless needed): The model burns extra energy parsing it.
For Example:
Choose smaller models for easy questions: Like GPT-3.5, Claude Instant
Local-first for simple tasks: Use small local models like LLaMA2 or GPT4All for calculations, basic summaries.
Disable Google’s AI suggestion as having it on probably 10x’s the energy usage, if not more. → reddit post of google's ai suggestion in regards to energy usage
→ How to disable google’s ai suggestion: https://tenbluelinks.org/
Note: Table provided above is not 100% accurate, i decided to just go with how I should approach my questions when asking chatgpt instead of trying to find a model that uses less energy. There were a lot of back & forth on how much each model uses in energy, so it’s more productive of me to make my questions in the future more clear and concise, and also turning google ai suggestion off.