Among the 40 models scoring Top tier (152+) across 58 benchmarks, the heaviest uses about 2790× the energy of the lightest, for work of the same standard. Choosing well costs you almost nothing in capability.
| Model | Maker | Capability | Energy / answer ↑ | Confidence | Thinking | Released |
|---|---|---|---|---|---|---|
| o3-mini | OpenAI | 140.3 | 0.248 Wh | Day-one estimate | Hidden | 2025-01-31 |
| Gemini 2.5 Flash-Lite (Jun 2025) | Google DeepMind | 133.9 | 0.249 Wh | Day-one estimate | Hidden | 2025-06-17 |
| o4-mini | OpenAI | 145.6 | 0.258 Wh | Day-one estimate | Hidden | 2025-04-16 |
| GPT-5 nano | OpenAI | 139.3 | 0.259 Wh | Day-one estimate | Hidden | 2025-08-07 |
| GPT-4o mini | OpenAI | 126.6 | 0.264 Wh | Day-one estimate | None | 2024-07-18 |
| stablelm-tuned-alpha-7b | 56.0 | 0.280 Wh | Inferred | Not stated | 2023-04-19 | |
| DeepSeek Coder 1.3B | DeepSeek +1 | 63.9 | 0.280 Wh | Inferred | Shown | 2023-11-02 |
| Qwen2.5-Coder-0.5B | 88.3 | 0.280 Wh | Inferred | Not stated | 2024-09-18 | |
| StarCoder 2 3B | Hugging Face +3 | 88.9 | 0.281 Wh | Inferred | Shown | 2024-02-22 |
| Gemma 3 1B | Google DeepMind | - | 0.282 Wh | Calculated | Shown | 2025-03-12 |
| Cerebras-GPT-13B | Cerebras Systems | 83.4 | 0.282 Wh | Inferred | Shown | 2023-03-20 |
| DeepSeek Coder 6.7B | DeepSeek +1 | 89.8 | 0.282 Wh | Inferred | Shown | 2023-11-02 |
| INTELLECT-1 | Prime Intellect +2 | 101.0 | 0.283 Wh | Inferred | Not stated | 2024-11-29 |
| CodeQwen1.5-7B | 95.0 | 0.283 Wh | Inferred | Not stated | 2024-04-15 | |
| Llama 3.2 1B | Meta AI | 102.0 | 0.283 Wh | Calculated | Shown | 2024-09-24 |
| StarCoder 2 7B | Hugging Face +3 | 93.7 | 0.283 Wh | Inferred | Shown | 2024-02-20 |
| Phi-1.5 | Microsoft | 91.7 | 0.283 Wh | Calculated | Shown | 2023-09-11 |
| Qwen-1_8B | 93.1 | 0.284 Wh | Inferred | Not stated | 2023-11-30 | |
| Qwen3-1.7B | Alibaba | - | 0.284 Wh | Calculated | Shown | 2025-04-29 |
| DeepSeek-R1-Distill-Qwen-1.5B | DeepSeek | - | 0.284 Wh | Calculated | Shown | 2025-01-20 |
| Grok 4.7 | xAI | 154.4 | 0.285 Wh | Calculated | Not stated | 2026-09-21 |
| Qwen2.5-Coder (1.5B) | Alibaba | 103.1 | 0.286 Wh | Inferred | Shown | 2024-09-18 |
| Baichuan1-7B | Baichuan | 90.6 | 0.286 Wh | Inferred | Shown | 2023-06-01 |
| Gemma 2B | Google DeepMind | 94.3 | 0.286 Wh | Calculated | Shown | 2024-02-21 |
| Dolly 2.0-12b | Databricks | 89.8 | 0.286 Wh | Inferred | Shown | 2023-04-11 |
| RedPajama-INCITE-7B-Base | 90.6 | 0.287 Wh | Inferred | Not stated | 2023-05-04 | |
| open_llama_7b | 91.8 | 0.287 Wh | Inferred | Not stated | 2023-06-07 | |
| Qwen3-30B-A3B | Alibaba | 136.2 | 0.287 Wh | Calculated | Shown | 2025-04-29 |
| DeepSeek Coder 33B | DeepSeek +1 | 96.5 | 0.288 Wh | Inferred | Shown | 2023-11-02 |
| Ministral 3B | Mistral AI | 118.1 | 0.288 Wh | Calculated | Not stated | 2024-10-16 |
| XGen-7B | Salesforce | 93.5 | 0.288 Wh | Inferred | Shown | 2023-06-27 |
| gpt-oss-20b | OpenAI | 137.8 | 0.289 Wh | Calculated | Shown | 2025-08-05 |
| Baichuan 2-7B | Baichuan | 96.5 | 0.289 Wh | Inferred | Shown | 2023-09-20 |
| Gemma 3 4B | Google DeepMind | 116.0 | 0.290 Wh | Calculated | Shown | 2025-03-12 |
| Qwen3-4B | Alibaba | - | 0.290 Wh | Calculated | Shown | 2025-04-29 |
| Qwen3.5-4B | Alibaba | - | 0.290 Wh | Calculated | Shown | 2026-02-24 |
| Qwen2.5-Coder-3B | 107.8 | 0.292 Wh | Inferred | Not stated | 2024-09-18 | |
| MPT-7B | MosaicML | 94.7 | 0.293 Wh | Inferred | Shown | 2023-05-05 |
| vicuna-13b-v1.1 | 94.8 | 0.294 Wh | Inferred | Not stated | 2023-04-12 | |
| Falcon-7B | Technology Innovation Institute | 95.2 | 0.295 Wh | Inferred | Shown | 2023-04-24 |
| Llama 2-7B | Meta AI | 99.2 | 0.298 Wh | Calculated | Shown | 2023-07-18 |
| Eurus-2-7B-PRIME | Tsinghua University +5 | - | 0.298 Wh | Calculated | Shown | 2024-12-31 |
| Mistral 7B | Mistral AI | - | 0.298 Wh | Calculated | Shown | 2024-05-27 |
| Mistral 7B v0.1 | Mistral AI | 112.3 | 0.298 Wh | Calculated | Shown | 2023-09-27 |
| Mistral 7B v0.3 | Mistral AI | 109.1 | 0.298 Wh | Calculated | Shown | 2024-05-27 |
| Qwen2.5-7B | Alibaba | 118.5 | 0.299 Wh | Calculated | Shown | 2024-09-19 |
| StarCoder 2 15B | Hugging Face +3 | 105.1 | 0.299 Wh | Inferred | Shown | 2024-02-20 |
| Llama 3.1-8B | Meta AI | 116.6 | 0.300 Wh | Calculated | Shown | 2024-07-23 |
| Llama 3-8B | Meta AI | 116.5 | 0.300 Wh | Calculated | Shown | 2024-04-18 |
| Gemini 1.5 Flash 8B | Google DeepMind | - | 0.300 Wh | Calculated | None | 2024-10-03 |
| Ministral 8B | Mistral AI | - | 0.300 Wh | Calculated | Shown | 2024-10-16 |
| Qwen3-8B | Alibaba | 136.2 | 0.301 Wh | Calculated | Shown | 2025-04-28 |
| chatglm2-6b | 99.0 | 0.301 Wh | Inferred | Not stated | 2023-06-24 | |
| Gemma 7B | Google DeepMind | 112.0 | 0.302 Wh | Calculated | Shown | 2024-02-21 |
| DeepSeek-Coder-V2-Lite-Base | 109.0 | 0.302 Wh | Inferred | Not stated | 2024-06-13 | |
| Qwen3.5-9B | Alibaba | 139.4 | 0.303 Wh | Calculated | Shown | 2026-02-24 |
| Gemma 2 9B | Google DeepMind | 119.8 | 0.303 Wh | Calculated | Shown | 2024-06-24 |
| LLaMA-7B | Meta AI | 96.8 | 0.304 Wh | Inferred | Shown | 2023-02-24 |
| Falcon 2 11B | Technology Innovation Institute | 109.6 | 0.307 Wh | Inferred | Shown | 2024-05-09 |
| Yi 6B | 01.AI | 104.9 | 0.307 Wh | Inferred | Shown | 2023-11-22 |
| MPT-30B | MosaicML | 100.8 | 0.307 Wh | Inferred | Shown | 2023-06-22 |
| Yi-9B | 107.7 | 0.308 Wh | Inferred | Not stated | 2024-03-01 | |
| Nemotron-4 15B | Nvidia | 107.8 | 0.309 Wh | Inferred | Not stated | 2024-02-26 |
| Baichuan2-13B | Baichuan | 103.3 | 0.310 Wh | Inferred | Shown | 2023-09-06 |
| Gemma 3 12B | Google DeepMind | 123.5 | 0.310 Wh | Calculated | Shown | 2025-03-12 |
| Mistral NeMo | Mistral AI | 118.7 | 0.310 Wh | Calculated | Shown | 2024-07-18 |
| Llama 2-13B | Meta AI | 106.2 | 0.313 Wh | Calculated | Shown | 2023-07-18 |
| Phi-4 | Microsoft Research | 130.4 | 0.316 Wh | Calculated | Shown | 2024-12-12 |
| phi-3-medium 14B | Microsoft | 121.2 | 0.316 Wh | Calculated | Shown | 2024-04-23 |
| internlm-7b | 103.0 | 0.317 Wh | Inferred | Not stated | 2023-07-05 | |
| Qwen3-14B | Alibaba | 138.2 | 0.318 Wh | Calculated | Shown | 2025-04-29 |
| DeepSeek-R1-Distill-Qwen-14B | DeepSeek | 135.4 | 0.318 Wh | Calculated | Shown | 2025-01-20 |
| Phi-2 | Microsoft | 108.0 | 0.319 Wh | Inferred | Shown | 2023-12-12 |
| Gemma 4 26B A4B | Google DeepMind | 141.8 | 0.339 Wh | Day-one estimate | Shown | 2026-04-02 |
| Magistral Small 1.0 | Mistral AI | 133.2 | 0.341 Wh | Calculated | Shown | 2025-06-10 |
| Mistral Small 3.2 | Mistral AI | 131.8 | 0.341 Wh | Calculated | Shown | 2025-06-20 |
| Magistral Small 1.2 | Mistral AI | 131.4 | 0.341 Wh | Calculated | Shown | 2025-09-18 |
| Mistral Small 3.1 | Mistral AI | 127.5 | 0.341 Wh | Calculated | Shown | 2025-03-17 |
| Mistral Small 3 | Mistral AI | 127.1 | 0.341 Wh | Calculated | Shown | 2025-01-30 |
| Gemma 3 27B | Google DeepMind | 130.1 | 0.349 Wh | Calculated | Shown | 2025-03-12 |
| Gemma 2 27B | Google DeepMind | 122.1 | 0.349 Wh | Calculated | Shown | 2024-06-24 |
| Qwen3.6 27B | Alibaba | 146.5 | 0.351 Wh | Calculated | Shown | 2026-04-22 |
| Gemma 4 31B IT | Google DeepMind | 142.9 | 0.359 Wh | Day-one estimate | Shown | 2026-04-02 |
| DeepSeek-R1-Distill-Qwen-32B | DeepSeek | 137.4 | 0.362 Wh | Calculated | Shown | 2025-01-20 |
| Qwen1.5-32B | Alibaba | - | 0.362 Wh | Calculated | Shown | 2024-04-03 |
| QwQ-32B | Alibaba | 137.6 | 0.363 Wh | Calculated | Shown | 2025-03-05 |
| Qwen2.5-32B | Alibaba | 128.5 | 0.363 Wh | Calculated | Shown | 2024-09-17 |
| Qwen3-32B | Alibaba | 138.5 | 0.364 Wh | Calculated | Shown | 2025-04-29 |
| Qwen2.5-Coder-14B | 116.4 | 0.448 Wh | Inferred | Not stated | 2024-09-18 | |
| LLaMA-13B | Meta AI | 100.7 | 0.459 Wh | Inferred | Shown | 2023-02-24 |
| Qwen-7B | Alibaba | 106.9 | 0.461 Wh | Inferred | Shown | 2023-09-28 |
| Llama 2-34B | Meta AI | 105.3 | 0.466 Wh | Inferred | Not stated | 2023-07-18 |
| Qwen2.5-Coder-32B | Alibaba | 119.5 | 0.469 Wh | Inferred | Shown | 2024-09-18 |
| Command R+ | Cohere +1 | 119.4 | 0.473 Wh | Inferred | Shown | 2024-08-30 |
| Falcon-40B | Technology Innovation Institute | 104.5 | 0.475 Wh | Inferred | Shown | 2023-05-25 |
| GPT-6 Luna | OpenAI | 156.3 | 0.492 Wh | Day-one estimate | Hidden | 2026-09-22 |
| PaLM 2-S | 106.3 | 0.495 Wh | Inferred | Not stated | 2023-05-17 | |
| Claude Haiku 5.5 | Anthropic | 159.0 | 0.515 Wh | Day-one estimate | Hidden | 2026-10-07 |
| Claude Haiku 4.5 | Anthropic | 142.4 | 0.550 Wh | Day-one estimate | Hidden | 2025-10-15 |
| GPT-4.1 nano | OpenAI | 129.6 | 0.551 Wh | Day-one estimate | Hidden | 2025-04-14 |
| Qwen 3.6 35B-A3B | Alibaba | 144.1 | 0.575 Wh | Calculated | Shown | 2026-04-14 |
| Qwen 3.5 Flash (hosted 35B-A3B) | Alibaba | 143.8 | 0.575 Wh | Calculated | Not stated | 2026-02-25 |
| Qwen3.5-35B-A3B | Alibaba | 142.5 | 0.575 Wh | Calculated | Shown | 2026-02-24 |
| Mixtral 8x7B | Mistral AI | 118.5 | 0.583 Wh | Day-one estimate | Shown | 2023-12-11 |
| GPT-5.4 Nano | OpenAI | 146.1 | 0.603 Wh | Day-one estimate | Hidden | 2026-03-17 |
| Amazon Nova Pro | Amazon | 123.8 | 0.687 Wh | Inferred | Not stated | 2024-12-03 |
| phi-3-mini 3.8B | Microsoft | 117.4 | 0.701 Wh | Inferred | Shown | 2024-04-23 |
| Gemini 1.5 Flash | Google DeepMind | 122.6 | 0.704 Wh | Inferred | None | 2024-09-24 |
| internlm-20b | 112.2 | 0.733 Wh | Inferred | Not stated | 2023-09-18 | |
| Yi-34B | 01.AI | 117.4 | 0.734 Wh | Calculated | Shown | 2023-11-22 |
| Yi-1.5-34B | 01.AI | - | 0.734 Wh | Calculated | Shown | 2024-05-13 |
| PaLM 2-M | 108.4 | 0.739 Wh | Inferred | Not stated | 2023-05-17 | |
| Falcon-180B | Technology Innovation Institute | 112.2 | 0.742 Wh | Inferred | Shown | 2023-09-06 |
| Claude 3 Haiku | Anthropic | 118.4 | 0.747 Wh | Inferred | None | 2024-03-07 |
| Qwen-14B | Alibaba | 113.1 | 0.748 Wh | Inferred | Shown | 2023-09-24 |
| LLaMA-33B | Meta AI | 107.5 | 0.778 Wh | Inferred | Shown | 2023-02-24 |
| Qwen3.7 Flash | Alibaba | 144.6 | 0.998 Wh | Inferred | Not stated | 2026-07-27 |
| GPT-5.5 Instant | OpenAI | 142.4 | 1.01 Wh | Inferred | Hidden | 2026-05-05 |
| Qwen 3.6 Flash | Alibaba | 143.1 | 1.03 Wh | Inferred | Not stated | 2026-04-27 |
| Claude 3.5 Haiku | Anthropic | 127.2 | 1.10 Wh | Inferred | None | 2024-10-22 |
| Gemini 1.0 Pro | Google DeepMind | 117.1 | 1.10 Wh | Inferred | None | 2023-12-13 |
| phi-3-small 7.4B | Microsoft | 121.8 | 1.12 Wh | Inferred | Shown | 2024-04-23 |
| Claude 3 Sonnet | Anthropic | 120.7 | 1.14 Wh | Inferred | None | 2024-02-29 |
| Grok-2 | xAI | 130.5 | 1.16 Wh | Inferred | Not stated | 2024-12-12 |
| gpt-oss-120b | OpenAI | 140.0 | 1.17 Wh | Calculated | Shown | 2025-08-05 |
| LLaMA-65B | Meta AI | 110.2 | 1.21 Wh | Inferred | Shown | 2023-02-24 |
| DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek | 124.8 | 1.22 Wh | Inferred | Shown | 2024-05-07 |
| Claude 2.1 | Anthropic | 119.3 | 1.22 Wh | Inferred | None | 2023-11-21 |
| Qwen3.5-122B-A10B | Alibaba | - | 1.22 Wh | Calculated | Shown | 2026-02-24 |
| Llama 4 Scout | Meta AI | 129.7 | 1.29 Wh | Calculated | Shown | 2025-04-05 |
| GPT-5 mini | OpenAI | 145.4 | 1.31 Wh | Day-one estimate | Hidden | 2025-08-07 |
| PaLM 2-L | 115.1 | 1.31 Wh | Inferred | Not stated | 2023-05-17 | |
| Stable Beluga 2 | Stability AI | 117.2 | 1.31 Wh | Inferred | Shown | 2023-07-20 |
| Gemini 3.5 Flash-Lite | Google DeepMind | 145.2 | 1.46 Wh | Day-one estimate | Hidden | 2026-07-21 |
| DBRX | Databricks | - | 1.49 Wh | Calculated | Shown | 2024-03-27 |
| GPT-3.5 Turbo | OpenAI | 113.4 | 1.56 Wh | Day-one estimate | None | 2024-01-25 |
| Qwen3.7-Plus | Alibaba | 147.3 | 1.60 Wh | Inferred | Not stated | 2026-06-02 |
| DeepSeek-V4-Flash | DeepSeek | 146.1 | 1.60 Wh | Inferred | Shown | 2026-04-24 |
| Qwen3-30B-A3B-Instruct (Jul 2025) | Alibaba | 137.4 | 1.61 Wh | Inferred | Shown | 2025-07-29 |
| GPT-4.1 mini | OpenAI | 135.0 | 1.61 Wh | Day-one estimate | Hidden | 2025-04-14 |
| Mistral Large 2 | Mistral AI | 127.5 | 1.66 Wh | Day-one estimate | Shown | 2024-11-18 |
| Qwen2.5-Max | Alibaba | 132.5 | 1.67 Wh | Inferred | Not stated | 2025-01-25 |
| Qwen3-235B-A22B-Instruct (Jul 2025) | Alibaba | 138.9 | 1.70 Wh | Inferred | Shown | 2025-07-25 |
| Gemini 3.1 Flash-Lite | 144.5 | 1.77 Wh | Day-one estimate | Hidden | 2026-03-03 | |
| DeepSeek LLM 67B | DeepSeek | 110.5 | 1.81 Wh | Calculated | Shown | 2023-11-29 |
| Llama 3.3 70B | Meta AI | 127.3 | 1.84 Wh | Calculated | Shown | 2024-12-06 |
| Llama 3.1-70B | Meta AI | 125.9 | 1.84 Wh | Calculated | Shown | 2024-07-23 |
| Llama 3-70B | Meta AI | 122.9 | 1.84 Wh | Calculated | Shown | 2024-04-18 |
| Llama 2-70B | Meta AI | 113.8 | 1.84 Wh | Calculated | Shown | 2023-07-18 |
| DeepSeek-R1-Distill-Llama-70B | DeepSeek | - | 1.84 Wh | Calculated | Shown | 2025-01-20 |
| Hermes 2 Theta Llama-3 70B | Nous Research +1 | - | 1.84 Wh | Calculated | Shown | 2024-06-20 |
| Tulu 3 (Tülu 3) 70B | Allen Institute for AI +1 | - | 1.84 Wh | Calculated | Shown | 2024-11-21 |
| Qwen1.5-72B | Alibaba | - | 1.86 Wh | Calculated | Shown | 2024-02-04 |
| Gemini 1.5 Pro | Google DeepMind | 126.9 | 1.86 Wh | Inferred | None | 2024-05-14 |
| Qwen2.5-72B | Alibaba | 129.0 | 1.86 Wh | Calculated | Shown | 2024-09-19 |
| Qwen2-72B | Alibaba | 125.3 | 1.86 Wh | Calculated | Shown | 2024-06-07 |
| Llama 3.2 90B | Meta AI | 125.5 | 2.03 Wh | Calculated | Shown | 2024-09-24 |
| Mistral Medium 3.5 | Mistral AI | 141.5 | 2.16 Wh | Day-one estimate | Shown | 2026-04-28 |
| Mistral Medium 3 | Mistral AI | 134.1 | 2.19 Wh | Day-one estimate | Not stated | 2025-05-07 |
| MiniMax-M2.7 | MiniMax | 145.9 | 2.34 Wh | Inferred | Shown | 2026-03-18 |
| Qwen 3.8 27B | Alibaba | 151.6 | 2.41 Wh | Inferred | Shown | 2026-08-14 |
| DeepSeek-V3.1 | DeepSeek | 139.9 | 2.41 Wh | Inferred | Shown | 2025-08-21 |
| Qwen 3.6 Plus | Alibaba | 147.5 | 2.44 Wh | Inferred | Not stated | 2026-03-31 |
| Qwen3-30B-A3B-Thinking (Jul 2025) | Alibaba | 139.6 | 2.45 Wh | Inferred | Shown | 2025-07-30 |
| Qwen 3.5 Plus (hosted 397B-A17B) | Alibaba | 146.3 | 2.46 Wh | Inferred | Not stated | 2026-02-16 |
| MiniMax-M2.5 | MiniMax | 146.7 | 2.51 Wh | Inferred | Shown | 2026-02-12 |
| WizardLM-2 8x22B | Microsoft | - | 2.52 Wh | Day-one estimate | Shown | 2024-04-15 |
| Gemini 2.0 Flash | Google DeepMind +1 | 134.7 | 2.53 Wh | Inferred | Not stated | 2025-02-05 |
| Kimi K2 (Jul 2025) | Moonshot | 140.1 | 2.56 Wh | Inferred | Shown | 2025-07-12 |
| Gemini 2.0 Pro | Google DeepMind | 135.1 | 2.57 Wh | Inferred | Hidden | 2025-02-05 |
| GPT-5.6 Luna | OpenAI | 156.3 | 2.61 Wh | Day-one estimate | Hidden | 2026-07-09 |
| Gemini 2.0 Flash Thinking | Google DeepMind +1 | 135.4 | 2.65 Wh | Inferred | Not stated | 2025-01-21 |
| GPT-4.5 | OpenAI | 136.7 | 2.67 Wh | Inferred | Hidden | 2025-02-27 |
| Qwen3-235B-A22B | Alibaba | 139.3 | 2.69 Wh | Calculated | Shown | 2025-04-29 |
| Claude 3.5 Sonnet | Anthropic | 130.0 | 2.86 Wh | Inferred | None | 2024-06-20 |
| Claude 3 Opus | Anthropic | 126.9 | 2.94 Wh | Inferred | None | 2024-02-29 |
| Claude Instant | Anthropic | 120.3 | 2.94 Wh | Inferred | None | 2023-08-09 |
| Claude 2 | Anthropic | 120.2 | 3.06 Wh | Inferred | None | 2023-07-11 |
| Qwen 3.6 Max (Preview) | Alibaba | 149.2 | 3.61 Wh | Inferred | Not stated | 2026-04-20 |
| GPT-5.4 Mini | OpenAI | 148.8 | 3.73 Wh | Day-one estimate | Hidden | 2026-03-17 |
| DeepSeek-V3.2 | DeepSeek | 146.1 | 3.83 Wh | Inferred | Shown | 2025-12-01 |
| Grok 4 Fast | xAI | 144.2 | 3.91 Wh | Inferred | Not stated | 2025-09-19 |
| DeepSeek-V3.2-Exp | DeepSeek | 145.0 | 4.00 Wh | Inferred | Shown | 2025-09-29 |
| Claude Sonnet 4 | Anthropic | 141.7 | 4.18 Wh | Inferred | Hidden | 2025-05-22 |
| Grok-3 mini | xAI | 140.3 | 4.19 Wh | Inferred | Not stated | 2025-04-09 |
| Inkling-Small | Thinking Machines | 150.2 | 4.97 Wh | Calculated | Shown | 2026-07-15 |
| Mixtral 8x22B | Mistral AI | 122.0 | 5.13 Wh | Calculated | Shown | 2024-04-17 |
| Qwen3.5 397B-A17B | Alibaba | 146.7 | 5.17 Wh | Calculated | Shown | 2026-02-13 |
| Llama 4 Maverick | Meta AI | 132.2 | 5.17 Wh | Calculated | Shown | 2025-04-05 |
| Qwen3.8 Max (0902) | Alibaba | 154.9 | 5.39 Wh | Inferred | Not stated | 2026-09-01 |
| Muse Spark 1.1 | Meta AI | 154.0 | 5.60 Wh | Inferred | Not stated | 2026-07-09 |
| Muse Spark 1.2 | Meta AI | 154.9 | 5.60 Wh | Inferred | Not stated | 2026-08-05 |
| Claude Sonnet 5.5 | Anthropic | 164.4 | 5.64 Wh | Day-one estimate | Hidden | 2026-09-28 |
| Claude Sonnet 5 | Anthropic | 156.4 | 5.78 Wh | Day-one estimate | Hidden | 2026-06-30 |
| GLM-4.5 | Z.ai (Zhipu AI) +1 | - | 5.79 Wh | Calculated | Shown | 2025-08-03 |
| GLM-4.6 | Z.ai (Zhipu AI) +1 | - | 5.79 Wh | Calculated | Shown | 2025-09-30 |
| GPT-5.1 | OpenAI | 149.8 | 5.87 Wh | Day-one estimate | Hidden | 2025-11-13 |
| Claude Opus 4.1 | Anthropic | 144.1 | 5.90 Wh | Inferred | Hidden | 2025-08-05 |
| Muse Spark | Meta AI | 152.1 | 5.90 Wh | Inferred | Not stated | 2026-04-08 |
| Qwen3-235B-A22B-Thinking (Jul 2025) | Alibaba | 143.8 | 5.92 Wh | Inferred | Shown | 2025-07-25 |
| GPT-5 | OpenAI | 150.0 | 6.04 Wh | Day-one estimate | Hidden | 2025-08-07 |
| Claude Opus 4 | Anthropic | 142.7 | 6.21 Wh | Inferred | Hidden | 2025-05-22 |
| GPT-4.1 | OpenAI | 136.8 | 6.21 Wh | Day-one estimate | Hidden | 2025-04-14 |
| GPT-6 Sol | OpenAI | 162.6 | 6.44 Wh | Day-one estimate | Hidden | 2026-09-22 |
| GPT-6.1 Sol | OpenAI | 166.3 | 6.64 Wh | Day-one estimate | Hidden | 2026-09-29 |
| Claude 3.7 Sonnet | Anthropic | 141.1 | 6.66 Wh | Inferred | Hidden | 2025-02-24 |
| o1-mini | OpenAI | 135.8 | 6.69 Wh | Inferred | Hidden | 2024-09-12 |
| o1-preview | OpenAI | 134.8 | 6.70 Wh | Day-one estimate | Hidden | 2024-09-12 |
| GPT-5.3 Codex | OpenAI | 156.8 | 6.81 Wh | Day-one estimate | Hidden | 2026-02-05 |
| GPT-5.2 | OpenAI | 153.5 | 6.83 Wh | Day-one estimate | Hidden | 2025-12-11 |
| GPT-4o | OpenAI | 128.7 | 6.92 Wh | Day-one estimate | None | 2024-08-06 |
| Gemini 2.5 Flash | Google DeepMind | 139.9 | 6.97 Wh | Day-one estimate | Hidden | 2025-06-17 |
| GPT-5.6 Terra | OpenAI | 159.4 | 7.22 Wh | Day-one estimate | Hidden | 2026-07-09 |
| Claude Sonnet 4.5 | Anthropic | 146.8 | 7.42 Wh | Day-one estimate | Hidden | 2025-09-29 |
| Claude Sonnet 4.6 | Anthropic | 152.4 | 7.48 Wh | Day-one estimate | Hidden | 2026-02-17 |
| GPT-5.4 | OpenAI | 156.7 | 7.53 Wh | Day-one estimate | Hidden | 2026-03-05 |
| Muse Spark 1.3 | Meta AI | 156.8 | 8.28 Wh | Inferred | Not stated | 2026-09-02 |
| Qwen3.7-Max | Alibaba | 153.8 | 8.50 Wh | Inferred | Not stated | 2026-05-19 |
| Llama 3.1-405B | Meta AI | 128.8 | 10.2 Wh | Day-one estimate | Shown | 2024-07-23 |
| GLM-5.2 | Z.ai (Zhipu AI) | 151.9 | 10.3 Wh | Calculated | Shown | 2026-06-16 |
| Gemini 3 Flash | Google DeepMind | 151.8 | 10.9 Wh | Day-one estimate | Hidden | 2025-12-17 |
| Mistral Large | Mistral AI | 122.0 | 11.0 Wh | Day-one estimate | Not stated | 2024-02-26 |
| Grok 4.20 | xAI | 152.1 | 11.3 Wh | Day-one estimate | Not stated | 2026-02-17 |
| Grok 4.3 Beta | xAI | 148.9 | 11.3 Wh | Day-one estimate | Not stated | 2026-04-17 |
| Kimi K2 Thinking | Moonshot | 146.0 | 11.6 Wh | Calculated | Shown | 2025-11-06 |
| Mistral Large 4 | Mistral AI | 146.9 | 11.9 Wh | Day-one estimate | Not stated | 2026-10-06 |
| DeepSeek-V3 | DeepSeek | 132.3 | 12.0 Wh | Calculated | Shown | 2024-12-26 |
| Nemotron 3 Ultra | Nvidia | 145.8 | 13.5 Wh | Calculated | Shown | 2026-06-04 |
| GLM-5 | Z.ai (Zhipu AI) | 145.8 | 14.6 Wh | Day-one estimate | Shown | 2026-02-11 |
| o3-pro | OpenAI | 147.4 | 15.7 Wh | Inferred | Hidden | 2025-06-10 |
| Gemini 3 Pro | Google DeepMind | 152.9 | 15.9 Wh | Inferred | Hidden | 2025-11-18 |
| DeepSeek V4 Flash 0731 | DeepSeek | 154.5 | 16.1 Wh | Calculated | Shown | 2026-07-31 |
| Qwen3-Max | Alibaba | 142.3 | 16.4 Wh | Day-one estimate | Not stated | 2025-09-24 |
| Claude Opus 4.8 | Anthropic | 158.3 | 16.7 Wh | Day-one estimate | Hidden | 2026-05-28 |
| Claude Opus 5 | Anthropic | 162.6 | 17.0 Wh | Day-one estimate | Hidden | 2026-07-24 |
| Claude Opus 5.5 | Anthropic | 166.8 | 17.0 Wh | Day-one estimate | Hidden | 2026-09-22 |
| o3 | OpenAI | 146.9 | 17.0 Wh | Inferred | Hidden | 2025-04-16 |
| Gemini 3.6 Flash | Google DeepMind | 154.2 | 17.2 Wh | Day-one estimate | Hidden | 2026-07-21 |
| GLM-5.3-Flash | Z.ai (Zhipu AI) | 153.2 | 17.6 Wh | Calculated | Shown | 2026-08-20 |
| Gemini 3.8 Flash | Google DeepMind | 157.2 | 17.6 Wh | Day-one estimate | Hidden | 2026-09-02 |
| Gemini 3.7 Flash | Google DeepMind | 157.2 | 17.7 Wh | Day-one estimate | Hidden | 2026-08-13 |
| Claude Opus 4.7 | Anthropic | 156.3 | 18.2 Wh | Day-one estimate | Hidden | 2026-04-16 |
| Claude Opus 4.6 | Anthropic | 155.3 | 18.7 Wh | Day-one estimate | Hidden | 2026-02-05 |
| Claude Opus 4.5 | Anthropic | 150.2 | 19.1 Wh | Day-one estimate | Hidden | 2025-11-24 |
| GLM-4.7 | Z.ai (Zhipu AI) | 143.8 | 19.1 Wh | Calculated | Shown | 2025-12-22 |
| Gemini 3.5 Flash | Google DeepMind | 154.4 | 19.7 Wh | Day-one estimate | Hidden | 2026-05-19 |
| GPT-5.5 | OpenAI | 158.9 | 21.1 Wh | Day-one estimate | Hidden | 2026-04-23 |
| GPT-5.6 Sol | OpenAI | 161.7 | 21.5 Wh | Day-one estimate | Hidden | 2026-07-09 |
| MiniMax-M3 | MiniMax | 147.1 | 22.0 Wh | Calculated | Shown | 2026-06-01 |
| GPT-4 Turbo (Nov 2023) | OpenAI | 126.5 | 23.2 Wh | Day-one estimate | None | 2024-01-25 |
| DeepSeek V4 Pro 0813 | DeepSeek | 155.3 | 25.9 Wh | Calculated | Shown | 2026-08-13 |
| DeepSeek-V4-Pro | DeepSeek | 149.2 | 25.9 Wh | Calculated | Shown | 2026-04-24 |
| GLM-5.1 | Z.ai (Zhipu AI) | 150.0 | 27.6 Wh | Day-one estimate | Shown | 2026-04-07 |
| Kimi K2.6 | Moonshot | 150.9 | 32.9 Wh | Day-one estimate | Shown | 2026-04-20 |
| Kimi K2.7 Code | Moonshot | 149.9 | 32.9 Wh | Day-one estimate | Shown | 2026-06-12 |
| Kimi K2.5 | Moonshot | 148.0 | 32.9 Wh | Day-one estimate | Shown | 2026-01-27 |
| Gemini 4 Argon | Google DeepMind | 164.6 | 34.7 Wh | Inferred | Hidden | 2026-09-30 |
| Grok 4.6 | xAI | 156.3 | 39.8 Wh | Day-one estimate | Not stated | 2026-08-12 |
| Grok 4.5 | xAI | 153.9 | 39.8 Wh | Day-one estimate | Not stated | 2026-07-08 |
| Gemini 3.1 Pro | Google DeepMind | 154.7 | 49.7 Wh | Day-one estimate | Hidden | 2026-02-19 |
| DeepSeek V4.1 Flash | DeepSeek | 155.6 | 54.2 Wh | Calculated | Shown | 2026-09-09 |
| Claude Fable 5 | Anthropic | 162.1 | 54.5 Wh | Day-one estimate | Hidden | 2026-06-09 |
| Claude Fable 5.1 | Anthropic | 164.7 | 54.9 Wh | Day-one estimate | Hidden | 2026-09-01 |
| DeepSeek-R1 | DeepSeek | 139.0 | 63.9 Wh | Calculated | Shown | 2025-01-20 |
| GPT-6 Astra | OpenAI | 166.4 | 64.0 Wh | Day-one estimate | Hidden | 2026-09-03 |
| Qwen 3.8 Max | Alibaba | 156.9 | 66.9 Wh | Calculated | Not stated | 2026-08-02 |
| GPT-4 (Mar 2023) | OpenAI | 123.2 | 67.5 Wh | Day-one estimate | None | 2023-03-14 |
| Kimi K3 | Moonshot | 157.4 | 69.9 Wh | Calculated | Shown | 2026-07-16 |
| GLM-5.3 | Z.ai (Zhipu AI) | 155.9 | 69.9 Wh | Calculated | Shown | 2026-08-14 |
| Gemini 2.5 Pro (Mar 2025) | Google DeepMind | 142.4 | 72.1 Wh | Day-one estimate | Hidden | 2025-03-31 |
| Inkling | Thinking Machines | 148.6 | 88.8 Wh | Calculated | Shown | 2026-07-15 |
| Grok 4 | xAI | 146.4 | 258 Wh | Day-one estimate | Not stated | 2025-07-09 |
| Grok 3 | xAI | 138.3 | 258 Wh | Day-one estimate | Not stated | 2025-04-09 |
| GPT-5 Pro | OpenAI | 150.7 | 265 Wh | Day-one estimate | Hidden | 2025-10-07 |
| GPT-5.2 Pro | OpenAI | 156.0 | 670 Wh | Day-one estimate | Hidden | 2025-12-11 |
| GPT-5.5 Pro | OpenAI | 162.0 | 730 Wh | Day-one estimate | Hidden | 2026-04-23 |
| GPT-5.4 Pro | OpenAI | 159.0 | 796 Wh | Day-one estimate | Hidden | 2026-03-05 |
| AI Co-Mathematician | Google DeepMind | - | - | No data | Hidden | 2026-05-08 |
| Gemini 2.5 Deep Think | Google +1 | - | - | No data | Not stated | 2025-08-01 |
| Grok 4 Heavy | xAI | - | - | No data | Not stated | 2025-07-10 |
| Muse Spark Open | Meta AI | - | - | No data | Shown | 2026-10-15 |
| Muse Wonder | Meta AI | - | - | No data | Not stated | 2026-10-12 |
| QWQ-Plus | Alibaba | - | - | No data | Not stated | 2025-04-08 |
| Qwen Plus | Alibaba | - | - | No data | Not stated | 2025-04-28 |
| Qwen-Turbo | Alibaba | - | - | No data | Not stated | 2024-11-01 |
Capability is Epoch AI's Epoch AI / Capabilities Index, a single score combining 58 separate tests, coding, using tools, holding a long document in mind, writing and science reasoning among them. It runs from roughly 55 to 170. We use the composite rather than any one test because a model can be excellent at science questions and mediocre at everything else. A handful of older models have no index score; we leave their capability blank rather than filling it from a different test, which would put them in the wrong place in the order.
Energy is calculated, not metered, no AI company publishes per-answer energy. We use the EcoLogits model, which works from how much of a model fires per word and how many words it writes: a 7,000-word-ish answer for a complex prompt, 300 for everyday chat, on a world-average electricity grid.
Calibration. The raw formula assumes a model has hardware largely to itself, which overstates real energy use. We corrected it against 46 samples vs ML.ENERGY on B200, batch 32-128, applying a factor of 0.53518.
Every model is calculated the same way, including those we have lab measurements for. Measured figures swing eightfold with how many requests a server handles at once, and only exist for open models, so using them directly would rank serving setups, not models. They are used to check our work instead.
Day-one estimates are models whose makers have never published a size. We bracket it from comparable models, so those bars are wide on purpose. The largest figures on this page carry the most uncertainty.
Hidden thinking. 65 of the models here generate reasoning text you never see. OpenAI, Anthropic and Google all confirm they bill for it while returning only a summary, or nothing. Our figures count a fixed answer length, so for those models the number shown is an understatement, not an estimate. Each one is marked in the table.
Sources. Epoch AI (capability, model sizes, roster), EcoLogits (energy model), ML.ENERGY (laboratory measurements). All public, all free.
Reuse. These figures are published under CC BY 4.0, use them anywhere, including commercially, as long as you credit The AI Footprint and link back. Read the full methodology →
Privacy. We count page visits in aggregate and set no cookies. What that means →