Ethics Series Post #4 — The Greener Choice: How Picking the Right AI Model Is an Act of Environmental Stewardship

I’ve been waiting for Earth Day to write about a topic that sits at the intersection of two things most nonprofit leaders care a lot about -- doing right by the environment and using technology to stretch limited resources in service of mission.

AI tools we've been promoting in this series have a real environmental cost. I wrote about this earlier. The energy and water consumption of large-scale AI infrastructure is significant, and the burden falls hardest on communities that are already vulnerable.

As a hiker, skier and outdoorswoman living in the Pacific Northwest, land stewardship is important to me. I just finished up a ski season in which very little snow fell on our nearby slopes. A visit last summer to North Cascade National Park exposed large swaths of browning evergreens—something I have never seen before. I also recreate on and around the Columbia River. It’s a river that I love and it is part of the lifeblood of my community. And it’s being used to cool the AI data centers in Washington. So, for me, environmental impacts are not theoretical. I see changes every day.

So far, at least among newbies (among whom I still count myself), there isn’t a lot of information out there about how to decrease the environmental impact of your AI usage. But the lack of info doesn’t mean there are not good choices out there. In fact, the choice of which AI model you use is itself an environmental decision. And for most of the tasks nonprofits do every day — writing thank you emails, drafting volunteer posts, cleaning up meeting minutes — you don't need the most powerful model available. You just need the right one for the job.

Energy Efficient AI Models are Available NOW

Each major AI product offers a lighter, more energy-efficient model. Typically, users don’t even know this is an option, or what they are using. Here are the lighter, more energy-efficient options available right now that are free or very low cost. They handle most nonprofit communications tasks with ease:

Claude Haiku — made by Anthropic The lightest model in the Claude family. Fast, efficient, and excellent for straightforward writing tasks like donor emails, social posts, meeting minutes, volunteer recruitment. Uses approximately 0.22 Wh per query. Available free at claude.ai. If you have a paid model, you can switch from the default (Sonnet) to Haiku in the text editor dropdown. Try it!

GPT-4o Mini — made by OpenAI A significantly lighter version of GPT-4o. Chat GPT will automatically use the Mini model for simple tasks, like the ones we typically do at nonprofits. Because it switches to the Mini model automatically based on your tasks, you can increase your chances that GPT will use Mini by asking it to “keep responses short and focused,” by minimizing the number of rewrites you ask for, and by thinking through your instructions before you ask for tasks to minimize complexity.

Gemini Flash — made by Google Google's lightweight model, designed for speed and efficiency. Gemini is also the default in the free version. Google reported in 2025 that the median Gemini prompt used 33 times less energy in May 2025 than it did a year earlier — a dramatic efficiency improvement. (Source: MIT Technology Review, citing Google, August 2025).

 How Efficient are the Lighter Models

I mostly use Claude, so I will start there. My paid account defaults to the “Sonnet” version, which you can see written in the text editor, but you can switch it to “Haiku” in the text editor, which is the much more energy efficient version. Claude also has an “Opus” version that would be used for complex tasks that I don’t normally engage in (coding, etc). Here is the difference between the lightest and heaviest models:

• Claude Haiku uses approximately 0.22 watt-hours per query and emits about 0.10 grams of CO2

• Claude Opus uses approximately 4.05 watt-hours, roughly 18 times more energy for the same task

A GPT Mini model response is equivalent to using your laptop or the LED light on your desk for a few seconds. In larger terms, 2000-10,000 GPT mini responses are equivalent to one hour of video streaming.

For routine nonprofit tasks — writing a donor email, cleaning up meeting notes, drafting a social post — a smaller model does the job just as well. The difference in output quality for everyday writing tasks is minimal. The difference in energy consumption is enormous.

An Option for Environmentally Focused Nonprofits – Run AI Locally

If you or your IT team is comfortable with it, running a model locally on your desktop eliminates the data transfer energy cost entirely and keeps your data on your own device. Tools like Ollama make this possible with small open-source models. More technical to set up — but the most privacy-protective and energy-efficient option available.

A Simple Framework

Think of it this way. You wouldn't drive a semi-truck to pick up groceries. The same logic applies to AI.

• Routine writing (emails, social posts, meeting minutes): Use Haiku, GPT-4o Mini, or Gemini Flash

• More complex tasks (grant narratives, strategic analysis): Use mid-tier models like Claude Sonnet or GPT-4o

• The most demanding work (long-form research, complex reasoning): Use large models sparingly and intentionally

Start with the lightest model. If the output isn't good enough, upgrade. You'll be surprised how often the smaller model delivers exactly what you need.

The Caveat

Individual nonprofit use of AI, even using large models every day, represents a genuinely small environmental footprint compared to the infrastructure-level energy consumption of big tech. The real environmental crisis is at the infrastructure level — the data centers, the power demands, the water consumption — and that requires policy change, not just individual behavior change.

But this still matters for nonprofits (and for me personally!)

Values alignment is central to who we are. When we make deliberate choices about which tools we use and how we use them, when we choose the lighter model because we can, we are practicing the same values-based decision-making that guides every other part of our work. It’s about baking integrity into every one of our actions. It’s about remaining conscious about our choices every day.

It's a small act. But it's a real one. And on Earth Day, take a moment to give it a try.

This is Post #4 in the Nonprofit AI Studio Ethics Series. Follow us on LinkedIn and Instagram @nonprofitaistudio so you never miss a post.