Sustainability & Carbon Tracking

Monitor your AI carbon footprint for ESG compliance and environmental responsibility.

Why Track Carbon?

✓ESG reporting requirements for enterprise organizations
✓Sustainability goals and carbon neutrality commitments
✓Understanding the true cost of AI beyond dollars
✓Making informed decisions about model selection

How Carbon Is Calculated

Every API request is tracked for its environmental impact using a two-factor model:

Energy Estimation

Based on model size and token count, we estimate the kWh consumed.

kWh = (tokens / 1,000,000) × model_energy_factor

Carbon Calculation

Using regional grid carbon intensity, we convert energy to CO2.

gCO2 = kWh × region_carbon_intensity

Model Energy Consumption

Different models have vastly different energy requirements:

Model	kWh per 1M tokens	Relative
Claude Haiku / GPT-4o-mini	0.005	1x
Llama 3.2 8B	0.008	1.6x
Claude Sonnet / GPT-4o	0.015	3x
DeepSeek V3	0.015	3x
Llama 70B / Qwen 72B	0.020	4x
Claude Opus / GPT-4	0.025	5x

Tip: Using smaller models for simple tasks can reduce your carbon footprint by 80% or more.

Regional Carbon Intensity

The carbon intensity of electricity varies dramatically by region:

EU North (Nordic)50 gCO2/kWh

Hydro & wind

US West (Oregon)120 gCO2/kWh

Hydro power

EU West (Ireland)290 gCO2/kWh

Mixed grid

US East (Virginia)380 gCO2/kWh

Mixed grid

US Central (Iowa)450 gCO2/kWh

Coal heavy

Asia East550 gCO2/kWh

High carbon

Routing to green regions can reduce carbon by up to 90%.

Understanding Your Impact

The dashboard shows your carbon footprint in relatable terms:

🚗

Car Miles

grams CO2 ÷ 400

1kg CO2 = 2.5 miles driven

🌳

Tree-Months

grams CO2 ÷ 1,750

1kg CO2 = 0.57 tree-months to offset

📱

Phone Charges

grams CO2 ÷ 8.22

1kg CO2 = 122 smartphone charges

Reducing Your Carbon Footprint

Use Smaller Models

Claude Haiku and GPT-4o-mini use 5x less energy than Opus/GPT-4 for many tasks.

Enable Green Routing

Route requests to providers in regions with renewable energy grids.

Batch Requests

Combining multiple small requests into one larger request reduces overhead.

Use Caching

Cached responses have zero carbon cost. Enable semantic caching for similar queries.

Host Open Source Models

Self-hosted models in green data centers can achieve 90%+ carbon reduction.

Sustainability Dashboard

Access your carbon metrics at Admin → Sustainability.

Summary Cards

Total CO2, carbon saved, energy used, and tokens processed

Environmental Equivalents

Car miles, tree-months, and phone charges visualization

Daily Charts

Carbon emissions over time with trend analysis

Provider Breakdown

Carbon by provider to identify optimization opportunities

API Reference

GET/api/admin/organizations/{orgId}/carbon

Returns carbon summary, daily data, and provider breakdown.

Query params: startDate, endDate