GPT-4o Image Generation Environmental Impact

Energy Consumption

GPT-4o's native image generation is distinct from DALL-E 3 — it uses autoregressive token generation rather than a separate diffusion model. Estimated at ~4.5 Wh per image, reflecting the full GPT-4o model generating high-resolution image tokens alongside text reasoning. This is more expensive than DALL-E 3 alone because the entire GPT-4o model is engaged.

Power Source & Carbon

Runs on Microsoft Azure, sharing infrastructure with GPT-4o text inference.

Water Usage

At approximately 16 mL per image, reflecting the extended GPU time required for autoregressive image token generation.

How much energy does GPT-4o Image Generation use per query?

Each GPT-4o Image Generation query consumes approximately 4.5 Wh of energy. This is 15x more than a traditional Google search (~0.3 Wh).

What is GPT-4o Image Generation's carbon footprint?

Based on the carbon intensity of Azure US East / Sweden, each query produces approximately 2.0 g of CO2. The grid in this region has a carbon intensity of 450 g CO2/kWh with 25% renewable energy.

How much water does GPT-4o Image Generation use?

Each query consumes approximately 16 mL of water, primarily used for cooling the data centers that process the request.

How does GPT-4o Image Generation compare to a Google search?

A GPT-4o Image Generation query uses 15x more than a Google search in terms of energy. A Google search uses approximately 0.3 Wh, while GPT-4o Image Generation uses 4.5 Wh.

Architecture

Autoregressive image generation (native to GPT-4o)

Release date

2025-03-25

Open source

No