Compare AI Models by Energy, CO₂ & Water

Text, Code & Chat Models

Gemini 2.0 Flash

0.08 Wh

~1 second of TV

GPT-4o Mini

0.14 Wh

~half a Google search

GPT-4.1 Mini

0.18 Wh

~half a Google search

LLaMA 3.2 1B

0.22 Wh

~half a Google search

Gemini 1.5 Pro

0.24 Wh

~half a Google search

GPT-4.1 Nano

0.27 Wh

~half a Google search

Gemini 2.5 Flash

0.35 Wh

~1 Google search

Phi-4

0.35 Wh

~1 Google search

GPT-4o

0.43 Wh

~1 Google search

GPT-4.1

0.45 Wh

~1 Google search

Grok 2

0.50 Wh

~1 Google search

Perplexity AI

0.50 Wh

~1 Google search

Claude 3.5 Haiku

0.55 Wh

~1 Google search

LLaMA 4 Scout

0.60 Wh

~charging a phone 10%

Gemma 3 27B

0.75 Wh

~charging a phone 10%

Gemini 2.5 Pro

0.90 Wh

~charging a phone 10%

LLaMA 3.1 70B

1.1 Wh

~charging a phone 10%

Grok 3

1.2 Wh

~charging a phone 10%

Qwen 3 235B

1.4 Wh

~charging a phone 10%

GitHub Copilot

1.5 Wh

~charging a phone 10%

Command R+

1.5 Wh

~charging a phone 10%

Mistral Large 2

1.8 Wh

~charging a phone 10%

LLaMA 4 Maverick

1.8 Wh

~charging a phone 10%

Claude Sonnet 4

2.5 Wh

~charging a phone 25%

Claude 3.7 Sonnet

2.8 Wh

~charging a phone 25%

DeepSeek-V3

4.2 Wh

~charging a phone 25%

Claude Opus 4

5.5 Wh

~a few phone charges

GPT-4.5

20.5 Wh

~running a TV for 10 min

o3 (Reasoning)

21.4 Wh

~running a TV for 10 min

DeepSeek-R1

29.0 Wh

~running a TV for 10 min

Image, Video & Music Generation

Imagen 3

2.5 Wh

~charging a phone 25%

Stable Diffusion XL

2.9 Wh

~charging a phone 25%

DALL-E 3

3.0 Wh

~charging a phone 25%

Adobe Firefly

3.2 Wh

~charging a phone 25%

Ideogram 2.0

3.5 Wh

~charging a phone 25%

Flux.1 Dev

3.8 Wh

~charging a phone 25%

GPT-4o Image Generation

4.5 Wh

~charging a phone 25%

Midjourney

7.5 Wh

~a few phone charges

Udio

70.0 Wh

~boiling a few kettles

Suno

75.0 Wh

~boiling a few kettles

Luma Dream Machine

160.0 Wh

~running a fridge for a day

Pika

180.0 Wh

~running a fridge for a day

Runway Gen-3 Alpha

200.0 Wh

~30 min of microwave

Kling

250.0 Wh

~30 min of microwave

Veo 2

350.0 Wh

~30 min of microwave

Sora

500.0 Wh

~30 min of microwave

Full Comparison

Model	Provider	Energy/Query	CO2/Query	Water/Query	Tier	Data Quality
Gemini 2.0 Flash	Google	0.08 Wh	0.01 g	0.09 mL	Ultra-efficient	Estimated
GPT-4o Mini	OpenAI	0.14 Wh	0.06 g	0.50 mL	Ultra-efficient	Estimated
GPT-4.1 Mini	OpenAI	0.18 Wh	0.08 g	0.65 mL	Ultra-efficient	Estimated
LLaMA 3.2 1B	Meta	0.22 Wh	0.08 g	1 mL	Ultra-efficient	Peer-reviewed
Gemini 1.5 Pro	Google	0.24 Wh	0.03 g	0.26 mL	Ultra-efficient	Verified
GPT-4.1 Nano	OpenAI	0.27 Wh	0.12 g	2 mL	Ultra-efficient	Peer-reviewed
Gemini 2.5 Flash	Google	0.35 Wh	0.05 g	0.38 mL	Ultra-efficient	Estimated
Phi-4	Microsoft	0.35 Wh	0.16 g	1 mL	Ultra-efficient	Estimated
GPT-4o	OpenAI	0.43 Wh	0.19 g	2 mL	Ultra-efficient	Peer-reviewed
GPT-4.1	OpenAI	0.45 Wh	0.20 g	2 mL	Ultra-efficient	Estimated
Grok 2	xAI	0.50 Wh	0.17 g	3 mL	Standard	Estimated
Perplexity AI	Perplexity	0.50 Wh	0.18 g	0.90 mL	Standard	Estimated
Claude 3.5 Haiku	Anthropic	0.55 Wh	0.19 g	1 mL	Standard	Estimated
LLaMA 4 Scout	Meta	0.60 Wh	0.24 g	1 mL	Standard	Estimated
Gemma 3 27B	Google	0.75 Wh	0.30 g	1 mL	Standard	Estimated
Gemini 2.5 Pro	Google	0.90 Wh	0.12 g	0.97 mL	Standard	Estimated
LLaMA 3.1 70B	Meta	1.1 Wh	0.44 g	2 mL	Standard	Peer-reviewed
Grok 3	xAI	1.2 Wh	0.46 g	2 mL	Standard	Estimated
Qwen 3 235B	Alibaba	1.4 Wh	0.77 g	5 mL	Standard	Estimated
GitHub Copilot	GitHub / OpenAI	1.5 Wh	1.0 g	10 mL	Standard	Estimated
Command R+	Cohere	1.5 Wh	0.53 g	3 mL	Standard	Estimated
Mistral Large 2	Mistral AI	1.8 Wh	1.1 g	45 mL	Standard	Verified
LLaMA 4 Maverick	Meta	1.8 Wh	0.72 g	3 mL	Standard	Estimated
Claude Sonnet 4	Anthropic	2.5 Wh	0.88 g	5 mL	Standard	Estimated
Imagen 3	Google	2.5 Wh	0.33 g	3 mL	Standard	Estimated
Claude 3.7 Sonnet	Anthropic	2.8 Wh	0.97 g	5 mL	Standard	Peer-reviewed
Stable Diffusion XL	Stability AI	2.9 Wh	1.2 g	5 mL	Standard	Peer-reviewed
DALL-E 3	OpenAI	3.0 Wh	1.2 g	15 mL	Heavy	Estimated
Adobe Firefly	Adobe	3.2 Wh	1.0 g	6 mL	Heavy	Estimated
Ideogram 2.0	Ideogram	3.5 Wh	1.2 g	7 mL	Heavy	Estimated
Flux.1 Dev	Black Forest Labs	3.8 Wh	1.5 g	7 mL	Heavy	Estimated
DeepSeek-V3	DeepSeek	4.2 Wh	2.3 g	16 mL	Heavy	Peer-reviewed
GPT-4o Image Generation	OpenAI	4.5 Wh	2.0 g	16 mL	Heavy	Estimated
Claude Opus 4	Anthropic	5.5 Wh	1.9 g	10 mL	Heavy	Estimated
Midjourney	Midjourney	7.5 Wh	2.3 g	14 mL	Heavy	Estimated
GPT-4.5	OpenAI	20.5 Wh	8.0 g	80 mL	Very heavy	Peer-reviewed
o3 (Reasoning)	OpenAI	21.4 Wh	10.0 g	100 mL	Very heavy	Peer-reviewed
DeepSeek-R1	DeepSeek	29.0 Wh	14.0 g	150 mL	Very heavy	Peer-reviewed
Udio	Udio	70.0 Wh	25.0 g	130 mL	Extreme	Estimated
Suno	Suno	75.0 Wh	30.0 g	350 mL	Extreme	Estimated
Luma Dream Machine	Luma AI	160.0 Wh	56.0 g	296 mL	Extreme	Estimated
Pika	Pika	180.0 Wh	63.0 g	330 mL	Extreme	Estimated
Runway Gen-3 Alpha	Runway	200.0 Wh	70.0 g	370 mL	Extreme	Estimated
Kling	Kuaishou	250.0 Wh	138.0 g	925 mL	Extreme	Estimated
Veo 2	Google	350.0 Wh	46.0 g	380 mL	Extreme	Estimated
Sora	OpenAI	500.0 Wh	225.0 g	1.85 L	Extreme	Estimated

Key Context

A traditional Google search uses ~0.3 Wh — comparable to the most efficient AI models.

Reasoning models (o3, DeepSeek-R1) use 20-100x more energy than standard chat models.

Image generation uses 5-40x more energy than text. One AI image = one full smartphone charge.

Video generation (Sora) uses ~500 Wh per 10-second clip — equivalent to running a microwave for about 30 minutes.

Efficiency is improving rapidly: Google achieved a 33x reduction in Gemini energy use in just 12 months.

Corporate emissions may be 7.6x higher than reported due to favorable accounting methods.