Model Environmental Impact Estimates

AI Model Environmental Leaderboard

Compare the environmental impact of popular AI models. Estimated CO₂e emissions, energy consumption, and overall rankings based on our calibrated environmental model.

Environmental Performance Rankings

Models ranked by overall environmental impact (lower emissions = better ranking)

Rank	Model	Provider	Parameters	CO₂e per 1M Tokens Input / Output	Energy per 1M Tokens Input / Output	Cost per 1M Tokens Input / Output
#1	moonshotai/kimi-vl-a3b-thinking	moonshotai	0.75B active (3B total)	~0.00006232 g (±25%)🟢 0.00003116 g/0.00009349 g	~119.9mWh (±25%) 59.9mWh/179.8mWh	$0.025/$0.10 Avg: $0.062
#2	meta-llama/llama-4-maverick	meta-llama	4.25B active (17B total)	~0.00013746 g (±25%)🟢 0.00006873 g/0.0002062 g	~624.8mWh (±25%) 312.4mWh/937.3mWh	$0.15/$0.60 Avg: $0.375
#3	meta-llama/llama-4-scout	meta-llama	4.25B active (17B total)	~0.00013746 g (±25%)🟢 0.00006873 g/0.0002062 g	~624.8mWh (±25%) 312.4mWh/937.3mWh	$0.08/$0.30 Avg: $0.19
#4	meta-llama/llama-3.1-8b-instruct	meta-llama	8B	~0.00022641 g (±25%)🟢 0.00011321 g/0.00033962 g	~1.0Wh (±25%) 514.6mWh/1.5Wh	$0.015/$0.02 Avg: $0.018
#5	z-ai/glm-4.5	z-ai	7B Est.	~0.00054732 g (±25%)🟢 0.00027366 g/0.00082098 g	~1.2Wh (±25%) 608.1mWh/1.8Wh	$0.20/$0.80 Avg: $0.50
#6	moonshotai/kimi-k2	moonshotai	7B Est.	~0.00058169 g (±25%)🟢 0.00029085 g/0.00087254 g	~1.1Wh (±25%) 559.3mWh/1.7Wh	$0.14/$2.49 Avg: $1.315
#7	qwen/qwen3-30b-a3b-instruct-2507	qwen	7.5B active (30B total)	~0.00062324 g (±25%)🟢 0.00031162 g/0.00093486 g	~1.2Wh (±25%) 599.3mWh/1.8Wh	$0.10/$0.30 Avg: $0.20
#8	openai/gpt-5-nano	openai	8B Est.	~0.000960 g (±25%)🟢 0.000480 g/0.00144 g	~3.4Wh (±25%) 1.7Wh/5.1Wh	$0.05/$0.40 Avg: $0.225
#9	mistralai/ministral-8b	mistralai	8B	~0.001955 g (±25%)🟢 0.00097736 g/0.002932 g	~4.3Wh (±25%) 2.2Wh/6.5Wh	$0.10/$0.10 Avg: $0.10
#10	meta-llama/llama-guard-4-12b	meta-llama	12B	~0.0399 g (±25%)🟡 0.020 g/0.0599 g	~181.5Wh (±25%) 90.7Wh/272.2Wh	$0.18/$0.18 Avg: $0.18
#11	anthropic/claude-3.5-haiku	anthropic	13B Est.	~0.0485 g (±25%)🟡 0.0242 g/0.0727 g	~186.5Wh (±25%) 93.3Wh/279.8Wh	$0.80/$4.00 Avg: $2.40
#12	google/gemini-2.0-flash-001	google	300B Est.	~0.0961 g (±25%)🟡 0.0481 g/0.1442 g	~801.2Wh (±25%) 400.6Wh/1201.8Wh	$0.10/$0.40 Avg: $0.25
#13	openai/o3-mini	openai	13B Est.	~0.1284 g (±25%)🟠 0.0642 g/0.1926 g	~458.5Wh (±25%) 229.2Wh/687.7Wh	$1.10/$4.40 Avg: $2.75
#14	mistralai/mistral-small-3.2-24b-instruct	mistralai	24B	~0.1351 g (±25%)🟠 0.0676 g/0.2027 g	~300.3Wh (±25%) 150.1Wh/450.4Wh	$0.05/$0.10 Avg: $0.075
#15	meta-llama/llama-3.3-70b-instruct	meta-llama	70B	~0.163 g (±25%)🟠 0.0815 g/0.2446 g	~741.1Wh (±25%) 370.5Wh/1111.6Wh	$0.038/$0.12 Avg: $0.079
#16	mistralai/codestral-2508	mistralai	22B Est.	~0.1769 g (±25%)🟠 0.0885 g/0.2654 g	~393.2Wh (±25%) 196.6Wh/589.8Wh	$0.30/$0.90 Avg: $0.60
#17	openai/gpt-5-mini	openai	20B Est.	~0.1975 g (±25%)🟠 0.0988 g/0.2963 g	~705.4Wh (±25%) 352.7Wh/1058.1Wh	$0.25/$2.00 Avg: $1.125
#18	deepseek/deepseek-chat-v3.1	deepseek	37B active (671B total)	~0.253 g (±25%)🟠 0.1265 g/0.3795 g	~486.6Wh (±25%) 243.3Wh/729.9Wh	$0.20/$0.80 Avg: $0.50
#19	z-ai/glm-4.5-air	z-ai	100B Est.	~0.2667 g (±25%)🟠 0.1333 g/0.400 g	~592.6Wh (±25%) 296.3Wh/888.9Wh	$0.20/$1.10 Avg: $0.65
#20	anthropic/claude-3.7-sonnet	anthropic	30B Est.	~0.2798 g (±25%)🟠 0.1399 g/0.4196 g	~1076.0Wh (±25%) 538.0Wh/1614.0Wh	$3.00/$15.00 Avg: $9.00
#21	openai/gpt-5-chat	openai	175B Est.	~0.2865 g (±25%)🟠 0.1432 g/0.4297 g	~1023.2Wh (±25%) 511.6Wh/1534.8Wh	$1.25/$10.00 Avg: $5.625
#22	qwen/qwen3-coder	qwen	120B active (480B total)	~0.3401 g (±25%)🟠 0.1701 g/0.5102 g	~654.1Wh (±25%) 327.0Wh/981.1Wh	$0.20/$0.80 Avg: $0.50
#23	google/gemini-2.5-flash	google	300B Est.	~0.3434 g (±25%)🟠 0.1717 g/0.5151 g	~2861.5Wh (±25%) 1430.8Wh/4292.3Wh	$0.30/$2.50 Avg: $1.40
#24	google/gemini-2.5-pro	google	300B Est.	~0.3434 g (±25%)🟠 0.1717 g/0.5151 g	~2861.5Wh (±25%) 1430.8Wh/4292.3Wh	$1.25/$10.00 Avg: $5.625
#25	qwen/qwen3-235b-a22b-2507	qwen	58.75B active (235B total)	~0.4018 g (±25%)🟠 0.2009 g/0.6026 g	~772.6Wh (±25%) 386.3Wh/1158.9Wh	$0.078/$0.312 Avg: $0.195
#26	qwen/qwen3-235b-a22b-thinking-2507	qwen	58.75B active (235B total)	~0.4018 g (±25%)🟠 0.2009 g/0.6026 g	~772.6Wh (±25%) 386.3Wh/1158.9Wh	$0.078/$0.312 Avg: $0.195
#27	baidu/ernie-4.5-300b-a47b	baidu	75B active (300B total)	~0.4826 g (±25%)🟠 0.2413 g/0.7239 g	~1072.4Wh (±25%) 536.2Wh/1608.6Wh	$0.28/$1.10 Avg: $0.69
#28	x-ai/grok-3-mini	x-ai	30B Est.	~0.6032 g (±25%)🟠 0.3016 g/0.9048 g	~1340.5Wh (±25%) 670.3Wh/2010.8Wh	$0.30/$0.50 Avg: $0.40
#29	x-ai/grok-4	x-ai	30B Est.	~0.6032 g (±25%)🟠 0.3016 g/0.9048 g	~1340.5Wh (±25%) 670.3Wh/2010.8Wh	$3.00/$15.00 Avg: $9.00
#30	anthropic/claude-opus-4	anthropic	400B Est.	~0.6184 g (±25%)🟠 0.3092 g/0.9276 g	~2378.4Wh (±25%) 1189.2Wh/3567.6Wh	$15.00/$75.00 Avg: $45.00
#31	x-ai/grok-3	x-ai	200B Est.	~0.6667 g (±25%)🟠 0.3333 g/1.0 g	~1481.5Wh (±25%) 740.8Wh/2222.3Wh	$3.00/$15.00 Avg: $9.00
#32	anthropic/claude-opus-4.1	anthropic	175B Est.	~0.6764 g (±25%)🟠 0.3382 g/1.01 g	~2601.4Wh (±25%) 1300.7Wh/3902.1Wh	$15.00/$75.00 Avg: $45.00
#33	openai/o3-pro	openai	175B Est.	~0.7162 g (±25%)🟠 0.3581 g/1.07 g	~2558.0Wh (±25%) 1279.0Wh/3837.0Wh	$20.00/$80.00 Avg: $50.00
#34	anthropic/claude-sonnet-4	anthropic	200B Est.	~0.773 g (±25%)🟠 0.3865 g/1.16 g	~2973.0Wh (±25%) 1486.5Wh/4459.5Wh	$3.00/$15.00 Avg: $9.00
#35	openai/gpt-5	openai	400B Est.	~1.64 g (±25%)🔴 0.8186 g/2.46 g	~5846.9Wh (±25%) 2923.4Wh/8770.3Wh	$1.25/$10.00 Avg: $5.625
#36	mistralai/mistral-medium-3.1	mistralai	500B Est.	~4.17 g (±25%)🔴 2.08 g/6.25 g	~9259.7Wh (±25%) 4629.8Wh/13889.5Wh	$0.40/$2.00 Avg: $1.20

Environmental Impact Comparison

CO₂e Impact Categories

Excellent

< 0.001g CO₂e/1M tokens

Good

0.001-0.01g CO₂e/1M tokens

Fair

0.01-0.1g CO₂e/1M tokens

Concerning

0.1-1g CO₂e/1M tokens

Poor

> 1g CO₂e/1M tokens

Real-World Comparisons

1M tokens ≈

• ~750 pages of text

• ~1.5 hours of conversation

• ~500 code files

Carbon Footprint Context

• 0.001g CO₂e ≈ 5mm car drive

• 0.1g CO₂e ≈ 50cm car drive

• 1g CO₂e ≈ 5m car drive

Legend

Estimation Indicators

Est.

Estimated using hybrid system (regex + Hugging Face API + cost data)

Parameters: Extracted from model names, API metadata, or cost-based estimation
Architecture: Detected using pattern matching and API data
Rank: Calculated from environmental metrics

Data updated: Just updated

±25% uncertainty range

Help us make your environmental data more accurate. We're constantly working to improve our measurements and would love to collaborate with you to ensure we're representing your models fairly and accurately.

Contact Our Research Team Learn About Our Methodology

Important Disclaimer

These environmental impact estimates are based on our calibrated models using publicly available data and research. Actual emissions may vary significantly based on hardware configuration, data center efficiency, energy sources, and usage patterns. We encourage providers to share more accurate data to improve these estimates. This data is provided for informational purposes and should not be considered as definitive environmental impact assessments.