How much energy does DeepSeek LLM 67B use?

DeepSeek · released 2023-11-29

DeepSeek LLM 67B uses about 0.077 Wh of electricity to answer a typical question. A long, complex answer costs about 23× more, at 1.81 Wh. That places it 145 of 277 among the models we can measure, and every figure here is calculated the same way as every other model.

A typical question

0.077 Wh

under 0.1 g CO₂·under a metre of driving

A long, complex answer

1.81 Wh

0.83 g CO₂·≈ 3 metres (11 feet) of driving

Capability

110.5

Energy rank

145 of 277

Human votes

1105

How capable is it, really?

The capability figure of 110.47 is a composite over 58 benchmarks, Epoch AI's Capabilities Index, which folds in software engineering, agentic tool use, long-context comprehension and creative writing as well as science reasoning. It is one number standing for a lot of separate tests, so treat a small gap between two models as noise rather than a ranking.

4,932 people have voted on it in blind head-to-head comparisons on LMArena, giving it a rating of 1105. This is the only figure here that measures whether anyone actually prefers the answers, rather than how the model scores on a test. Broken down, people rate it 1095 on coding, 1067 on creative writing, 1079 on following instructions.

What this figure is

Calculated Calculated from a published parameter count. Every model on this site is calculated the same way, so the comparison holds even where the absolute value carries uncertainty. Our figures are good to roughly a factor of two to three, which is why we only ever state a difference of 5× or more.

Is there a better choice?

A lighter model that does the job

Grok 4.7

xAI · capability 154.4. It scores higher than DeepSeek LLM 67B on Epoch's Capabilities Index (154.37 against 110.47), the composite over 58 benchmarks that the capability figure above comes from, and uses about 6.3× less energy per answer: 0.285 Wh against 1.81 Wh. It has not faced human head-to-head voting, where DeepSeek LLM 67B carries 4,932 votes, worth weighing.

What DeepSeek publishes

Energy cannot be worked out without knowing how big a model is. Of the 16 models from DeepSeek that we track, 11 can be assessed at all, the rest publish too little for anyone outside the company to calculate anything.

Where these figures come from

Capability and the roster come from Epoch AI, whose Capabilities Index is the 58-benchmark composite quoted above. Energy is calculated with EcoLogits and calibrated against laboratory measurements from ML.ENERGY. Human preference comes from LMArena. All four are public and free; the full method is here.

These figures are published under CC BY 4.0, use them anywhere, including commercially, with credit to The AI Footprint.

Compare it with something

Compare DeepSeek LLM 67B with another model

Or work out what a year of using it adds up to, for you or for a whole organisation.

Figure calculated 2026-10-17 · how these numbers are made · the full comparison of 277 models