When to use CodeCarbon vs EcoLogits
- CodeCarbon
- EcoLogits
AI and software need electricity. Where your code runs — on your laptop or on a provider’s servers — changes what you can measure directly.
The mlco2 ecosystem offers two main tools. They are complementary, not competing.
Quick answer
| CodeCarbon | EcoLogits | |
|---|---|---|
| What it tracks | Emissions from local computing — code on hardware you control | Emissions from GenAI API calls — OpenAI, Anthropic, Mistral, etc. |
| Typical use cases | Model training, local inference, scripts on your machine or cloud VM | Chatbots, agents, RAG pipelines calling remote models |
| How it works | Measures CPU, GPU & RAM power, applies regional grid carbon intensity | Estimates impacts from request metadata (model, tokens, latency) via LCA models |
| Get started | pip install codecarbon | pip install ecologits |
Use CodeCarbon when…
You run code on hardware you control or provision:
- Training or fine-tuning ML models on your laptop, workstation, or cluster
- Running batch jobs on a cloud VM you spin up
- Comparing experiments by CPU/GPU power and regional grid intensity
- Wrapping existing scripts with the CLI — no code changes required
CodeCarbon focuses on CPU, GPU, and RAM. It does not separately model disk I/O, network transfers, or cooling — those are usually much smaller for local code-level experiments.
from codecarbon import EmissionsTracker
tracker = EmissionsTracker()
tracker.start()
# your training or inference code
emissions = tracker.stop()Use EcoLogits when…
You call third-party GenAI APIs and want per-request impact estimates:
- Building chatbots, agents, or copilots on OpenAI, Anthropic, Mistral, etc.
- Running RAG pipelines where inference happens on a provider’s servers
- Reporting sustainability metrics for GenAI usage to stakeholders
- Comparing models by environmental cost per request
EcoLogits attaches impact data to responses from official provider SDKs:
from ecologits import EcoLogits
from openai import OpenAI
EcoLogits.init(providers=["openai"])
response = OpenAI().chat.completions.create(
model="gpt-4o-mini",
messages=[{"role": "user", "content": "Hello!"}],
)
print(response.impacts.gwp.value.mean, "kgCO2eq")Use both when your stack mixes them
Many teams train or fine-tune locally with CodeCarbon, then serve features through GenAI APIs tracked with EcoLogits. Same mission — measure what you can, where it actually runs.
No code? Try these first
| Tool | Best for |
|---|---|
| EcoLogits Calculator | Quick browser estimates for GenAI usage — no install |
| EcoLogits API | HTTP access to the same methodology — powers custom dashboards and editor plugins |
| CodeCarbon Dashboard | Visualise local experiment emissions over time |
Still unsure?
- You own the machine running the code → CodeCarbon
- You send a request to OpenAI / Anthropic / similar → EcoLogits
- You want a quick what-if without writing code → Calculator