3.0× less energy than Claude Sonnet 4.
Claude Sonnet 4 Environmental Impact
Anthropic's latest balanced model
The figures below are for one quick question. A longer chat with follow-ups uses about 0.73 Wh, and an hour or two of agent work about 21.0 Wh.
One query, in context
Energy, carbon and water for a single use. Quick single question (one exchange)
70% of a Google search.
62% of a Google search.
About 7.8× a drop of water and 8% of a teaspoon of water.
- Where it runsAWS US East (Virginia) / US West (Oregon)
- Grid carbon intensity450 g CO₂/kWh · 25% renewable
- Compared with a Google searchabout the same as a search (0.3 Wh)
- Water~2,564 queries to fill 1 litre
Where Claude Sonnet 4 sits
#58 of 199 by energy per query. Lighter than 72% of all models. Lit dots are the same kind of model; click any dot to open it.
Want something lighter?
3.5× less energy than Claude Sonnet 4.
11× less energy than Claude Sonnet 4.
What a month of use adds up to
Pick how much you use it. Based on one quick use at a time, five days a week.
- Energy
- 22.7 Wh
- Carbon
- 8.0 g
- Water
- 42 mL
A month of 5 chats a day, 5 days a week. The energy is about 1.9× a phone charge and 21% of a kettle boil.
Work out your own →- Energy
- 90.9 Wh
- Carbon
- 32.0 g
- Water
- 169 mL
A month of 20 chats a day, 5 days a week. The energy is about 7.6× a phone charge and 83% of a kettle boil.
Work out your own →- Energy
- 454.6 Wh
- Carbon
- 160.2 g
- Water
- 844 mL
A month of 100 chats a day, 5 days a week. The energy is about 4.1× a kettle boil and 6% of a UK home's day.
Work out your own →How does Claude Sonnet 4 compare?
Next to similar models. Bars use a log scale, so each step is a big jump.
Detailed breakdown
Energy
We estimate Claude Sonnet 4 at ~0.21 Wh for a quick single question, taken from the CLEER dataset (Sustainable AI Group, Sept 2026). The figure covers the full facility — accelerators, host servers, idle capacity and cooling. CLEER cannot measure closed models directly: it projects per-token energy from open models it has benchmarked, on B200/B300-class accelerators (proxy-match confidence 0.9 on a 0-1 scale). On the same basis a typical multi-turn chat comes to ~0.74 Wh and a typical agentic coding session to ~26 Wh, so how the model is used matters far more than which prompt you send. This replaces our earlier ~2.5 Wh estimate, which was extrapolated from 2025 benchmark figures on older hardware. Anthropic has published no per-query figures for this model.
Power source & carbon
Runs on AWS infrastructure in US East (Virginia) and US West (Oregon). Amazon claims to be the largest corporate renewable energy purchaser for 5 consecutive years. Anthropic has not published emissions data.
Water use
At ~0.39 mL per quick question, derived from the energy estimate at the ~1.9 L/kWh water factor we apply to this provider's infrastructure. Water scales directly with energy, so longer chats and agentic sessions use proportionally more.
About Claude Sonnet 4
Anthropic positions Claude as the safety-focused alternative to GPT and Gemini, but safety and sustainability are not the same thing. Claude Sonnet 4 runs primarily on AWS and Google Cloud infrastructure, inheriting the carbon intensity of those providers' data centre regions. Anthropic does not publish per-query energy data, so our estimate is derived from hardware benchmarks and API latency measurements. What distinguishes Claude in the efficiency conversation is Anthropic's constitutional AI approach — the model is trained to be helpful with shorter, more direct responses, which in practice means fewer output tokens and slightly lower energy per typical interaction.
These figures are estimates derived from hardware specifications and API benchmarks — Anthropic has not published official energy data for Claude Sonnet 4. Actual consumption may vary significantly depending on batching, quantisation, and infrastructure optimisations that we cannot observe from outside.
Claude Sonnet 4 in context
The Anthropic Claude family
One chart per tier, on a shared time axis. Hover a dot for details; click to open that version.
Newer in this tier: Claude Sonnet 5.5 uses 0.24 Wh per query, 14% more than Claude Sonnet 4.
22 versions across 4 tiers. Lightest: Claude Haiku 5.5 (0.02 Wh); heaviest: Claude Code (2.0 Wh). Charts share a time axis but each tier has its own energy scale, so its trend is clear; the percentage is the change from a tier’s first version to its latest.
What does your Claude Sonnet 4 usage cost the planet?
Use the calculator to estimate your personal footprint based on how often you use Claude Sonnet 4.
Frequently asked questions
How much energy does Claude Sonnet 4 use per query?
Each Claude Sonnet 4 query consumes approximately 0.21 Wh of energy. This is about the same as a traditional Google search (~0.3 Wh).
What is Claude Sonnet 4's carbon footprint?
Based on the carbon intensity of AWS US East (Virginia) / US West (Oregon), each query produces approximately 0.07 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 Claude Sonnet 4 use?
Each query consumes approximately 0.39 mL of water, primarily used for cooling the data centers that process the request.
How does Claude Sonnet 4 compare to a Google search?
A Claude Sonnet 4 query uses about the same as a Google search in terms of energy. A Google search uses approximately 0.3 Wh, while Claude Sonnet 4 uses 0.21 Wh.
Technical details
- Architecture
- Dense Transformer (decoder-only)
- Context window
- 200,000 tokens
- Release date
- 2025-05-22
- Open source
- No