In short
Use a reasoning or "thinking" mode for genuinely hard problems, such as multi-step maths, tricky code or careful analysis, and a standard or fast model for everyday questions, drafting and summaries. Reasoning modes write long hidden chains of working before answering, and one benchmark found they used on average 30 times more energy.
Steps
- Start with the standard or fast model for everyday questions and writing.
- Switch to a reasoning mode when the task needs several logical steps or careful checking.
- Turn reasoning off again once the hard part is done.
- Check whether your app chooses the mode automatically, and override it for simple tasks if you can.
Reasoning modes are for hard problems, not for every question. A "thinking" model works through a problem step by step before it answers, writing out many hidden tokens of working. That can improve answers on difficult tasks, but it costs far more computation.
How much more energy does reasoning use?
A lot more, on the evidence so far. Hugging Face's AI Energy Score team tested open models in late 2025 and found that reasoning models used, on average, 30 times more energy. When the same model was run with reasoning switched on and off, it used between 150 and 700 times more energy with reasoning enabled, and produced between 300 and 800 times more tokens. These are tests of open models on standard tasks; commercial services do not publish comparable figures.
Kim, Shin, Chung and Rhu found a related pattern for "test-time scaling" more generally: spending more computation per answer gave diminishing returns in accuracy.
When is it worth it?
| Task | Usually enough | Reasoning helps |
|---|---|---|
| Quick facts, rewording, emails | Standard or fast model | Rarely |
| Summaries and first drafts | Standard model | Rarely |
| Multi-step maths or logic | Sometimes | Often |
| Debugging tricky code | Sometimes | Often |
| Weighing a complex decision | Standard model to start | When the first answer is shallow |
Common mistakes
- Leaving reasoning on all the time because it feels "smarter".
- Using it for creative writing or simple rewrites, where it rarely helps.
- Not checking automatic mode-switching. Some apps decide for you; see whether you can choose.
For choosing models more generally, see choose a smaller AI model and choose the right model.
Sources
- AI Energy Score v2: refreshed leaderboard, now with reasoning, Luccioni and Gamazaychikov, Hugging Face (2025)
- The cost of dynamic reasoning: demystifying AI agents and test-time scaling from an AI infrastructure perspective, Kim, Shin, Chung and Rhu, HPCA 2026 (2025)
Related
How do I choose a smaller AI model?
How to pick the smallest AI model that does your task well: set a quality bar, test small to large, check energy ratings and keep a big model as fallback.
How do I use AI agents and deep research tools without waste?
When AI agents and deep research modes are worth their much higher token use, and how to brief them so they finish quickly instead of looping.
How do I reduce the energy my ChatGPT use consumes?
Practical ways to cut the energy of using ChatGPT and similar chatbots: shorter exchanges, fewer regenerations, lighter models and less image generation.
Choose the right model for the job
How to right-size AI: when to use small or specialised models, when a frontier model earns its cost, how routing works, and when not to use AI.