11× less energy than o3 (Reasoning).
o3 (Reasoning) Environmental Impact
Advanced reasoning model — very high energy use
The figures below are for one quick question. A longer chat with follow-ups uses about 8.0 Wh, and an hour or two of agent work about 230.0 Wh.
One query, in context
Energy, carbon and water for a single use. Quick single question (one exchange)
About 2.8× a minute of laptop use and 19% of a phone charge.
About 9.2× a Google search and 52% of 10 m by car.
About 1.7× a teaspoon of water and 3% of a glass of water.
- Where it runsAzure US East / Sweden
- Grid carbon intensity450 g CO₂/kWh · 25% renewable
- Compared with a Google search8x more than a search (0.3 Wh)
- Water~120 queries to fill 1 litre
Where o3 (Reasoning) sits
#155 of 199 by energy per query. Heavier than 77% of all models. Lit dots are the same kind of model; click any dot to open it.
Want something lighter?
3.8× less energy than o3 (Reasoning).
5.7× less energy than o3 (Reasoning).
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
- 249.0 Wh
- Carbon
- 119.1 g
- Water
- 898 mL
A month of 5 chats a day, 5 days a week. The energy is about 2.3× a kettle boil and 3% of a UK home's day.
Work out your own →- Energy
- 995.9 Wh
- Carbon
- 476.3 g
- Water
- 3.59 L
A month of 20 chats a day, 5 days a week. The energy is about 9.1× a kettle boil and 13% of a UK home's day.
Work out your own →- Energy
- 4.98 kWh
- Carbon
- 2.38 kg
- Water
- 17.97 L
A month of 100 chats a day, 5 days a week. The energy is about 45× a kettle boil and 67% of a UK home's day.
Work out your own →How does o3 (Reasoning) compare?
Next to similar models. Bars use a log scale, so each step is a big jump.
Detailed breakdown
Energy
We estimate o3 (Reasoning) at ~2.3 Wh for a quick single question. o3 is a reasoning-only model: it reasons before every answer, so we start from the standard-chat figure for its model class (GPT-4o, ~0.46 Wh on the CLEER basis (Sustainable AI Group, Sept 2026: full-facility energy for one quick question on current hardware)) and multiply by 5. That multiplier sits between the roughly 2x measured for medium reasoning effort and the 13x the Joule 2026 production study found for long reasoning-style responses; a quick question does not trigger the longest chains, but a hard problem can cost ten times more again. This is our extrapolation, not a CLEER figure or a measurement. The peer-reviewed benchmark this replaces as our headline — ~21.414 Wh for a medium reasoning query (~1000 tokens), from Jegham et al. (2025) — measured a heavier prompt class on 2025 hardware and remains a reasonable upper bound.
Power source & carbon
Runs on Microsoft Azure's infrastructure, primarily in the US East region. The Virginia data center corridor, where many Azure facilities are concentrated, has a carbon intensity of approximately 450 g CO2/kWh — significantly above the US average of 380 g CO2/kWh — due to reliance on natural gas and remaining coal generation.
Water use
At ~8.3 mL per quick question, derived from the energy estimate at the ~3.6 L/kWh water factor we apply to this model's infrastructure. Water scales directly with energy, so longer chats, reasoning and agentic sessions use proportionally more.
About o3 (Reasoning)
o3 represents a fundamentally different kind of AI energy consumption. Unlike standard chat models that respond in under a second, o3 'thinks' — running extended chains of internal reasoning that can take 30 seconds or more. That deliberation is what makes it capable of PhD-level maths and complex code generation, but it comes at a several-fold energy premium over models like GPT-4o on even a quick question, and far more on the hard problems it is built for. It is the clearest example of a tradeoff that will define AI sustainability: the most capable models are often the most expensive to run, and the gap is widening.
These figures are estimates derived from hardware specifications and API benchmarks — OpenAI has not published official energy data for o3 (Reasoning). Actual consumption may vary significantly depending on batching, quantisation, and infrastructure optimisations that we cannot observe from outside.
o3 (Reasoning) in context
Key insights
- Uses 6x more energy than the category average — reasoning models are inherently compute-intensive
The OpenAI o3 family
One chart per tier, on a shared time axis. Hover a dot for details; click to open that version.
What does your o3 (Reasoning) usage cost the planet?
Use the calculator to estimate your personal footprint based on how often you use o3 (Reasoning).
Frequently asked questions
How much energy does o3 (Reasoning) use per query?
Each o3 (Reasoning) query consumes approximately 2.3 Wh of energy. This is 8x more than a traditional Google search (~0.3 Wh).
What is o3 (Reasoning)'s carbon footprint?
Based on the carbon intensity of Azure US East / Sweden, each query produces approximately 1.1 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 o3 (Reasoning) use?
Each query consumes approximately 8 mL of water, primarily used for cooling the data centers that process the request.
How does o3 (Reasoning) compare to a Google search?
A o3 (Reasoning) query uses 8x more than a Google search in terms of energy. A Google search uses approximately 0.3 Wh, while o3 (Reasoning) uses 2.3 Wh.
Technical details
- Architecture
- Transformer with chain-of-thought reasoning (decoder-only)
- Context window
- 200,000 tokens
- Release date
- 2025-01-31
- Open source
- No
- Training data cutoff
- 2024-10
Sources
- Sustainable AI Group — CLEER dashboard (CLEER-Text-0926, Sep 2026)
- OpenAI — Introducing o3 and o3-mini (Jan 2025)
- arXiv — OpenAI o3 System Card (2025)
- Epoch AI — How much energy does ChatGPT use?
- arXiv — Benchmarking Energy, Water, and Carbon of LLM Inference (2025)
- Harvard T.H. Chan — Data center electricity carbon intensity 48% above US average