Reasoning Models
Also written as Reasoning Model, Thinking Models, Extended Thinking, Test-Time Compute, Inference-Time Scaling
AI models that 'think' before answering: they work through a problem step by step, check their own work and try alternatives, then give a final answer. Spending more computing time while answering (test-time compute) makes them much better at maths, coding and multi-step planning, but slower and more expensive per answer. OpenAI's o-series started the category in 2024; most major models now have a thinking mode.
Think of it like
The difference between blurting out an answer and taking a minute with a notepad first.
Junior or senior?
Using one is common; knowing when not to is the skill.
Senior sounds like
Can describe how they balanced thinking time against speed and cost for a real feature, and which tasks didn't benefit at all.
Ask them
“Which tasks in your product actually needed a reasoning model, and how did you control the extra cost and delay?”