Secondary reference

AI model reference library

Specs, pricing and a best-effort environmental estimate for each model. Figures per single text prompt, shown as a low–high band.

Footprint numbers are rough public estimates that vary by 10–100× across sources and depend on data-centre efficiency and local grid mix. They are shown as a low–high band with a source and date. Don't treat them as measured fact.

Anthropic

Claude Haiku 4.5

200k context · in $1 / out $5 per 1M

Energy ≈ 0.3 WhWater ≈ 0.35 mLCO₂e ≈ 0.12 g

Claude Opus 5

200k context · in $15 / out $75 per 1M

Energy ≈ 3 WhWater ≈ 3.3 mLCO₂e ≈ 1.3 g

Claude Sonnet 5

200k context · in $3 / out $15 per 1M

Energy ≈ 1 WhWater ≈ 1.1 mLCO₂e ≈ 0.45 g

DeepSeek

DeepSeek-V3

128k context · in $0.27 / out $1.1 per 1M

Energy ≈ 1 WhWater ≈ 1.1 mLCO₂e ≈ 0.6 g

Google

Gemini 2.5 Flash

1,049k context · in $0.3 / out $2.5 per 1M

Energy ≈ 0.24 WhWater ≈ 0.26 mLCO₂e ≈ 0.03 g

Gemini 2.5 Pro

1,049k context · in $1.25 / out $10 per 1M

Energy ≈ 0.9 WhWater ≈ 1 mLCO₂e ≈ 0.12 g

Mistral

Mistral Large 2

128k context · in $2 / out $6 per 1M

Energy ≈ 1.1 WhWater ≈ 1.2 mLCO₂e ≈ 1.4 g

OpenAI

GPT-4o

128k context · in $2.5 / out $10 per 1M

Energy ≈ 0.3 WhWater ≈ 0.45 mLCO₂e ≈ 0.2 g

GPT-5

400k context · in $1.25 / out $10 per 1M

Energy ≈ 1.8 WhWater ≈ 2 mLCO₂e ≈ 0.8 g

200k context · in $2 / out $8 per 1M

Energy ≈ 3.5 WhWater ≈ 4 mLCO₂e ≈ 1.6 g

xAI

Grok 4

256k context · in $3 / out $15 per 1M

Energy ≈ 2.5 WhWater ≈ 2.8 mLCO₂e ≈ 1.9 g