A category of AI model designed to reason through problems step by step before producing an answer — trading speed for greater accuracy on complex tasks like multi-step strategy, analysis, and nuanced judgment.
Also known as reasoning model, thinking model
Standard large language models generate responses by predicting what text should come next, based on patterns learned from training data. They are fast and capable on a wide range of tasks. Large Reasoning Models take a different approach: before producing an answer, they work through the problem — considering alternatives, testing intermediate steps, and arriving at a conclusion through deliberate reasoning rather than pattern completion.
The trade-off is speed and cost. LRMs take longer to respond and cost more to run than standard LLMs on the same task. The benefit is accuracy on problems that require multiple steps, nuanced judgment, or resistance to obvious but wrong answers. Examples include OpenAI’s o-series models and similar thinking-mode models from other providers.
Vendors are increasingly marketing “reasoning mode” as a feature. Agencies advising clients on AI adoption need to understand what that means: a model that takes longer and costs more but handles complexity better. Knowing when that trade-off is worth it is becoming a core AI literacy skill for strategy and technology teams.
Not every task benefits from reasoning. Writing a social caption does not require an LRM. Analyzing attribution data to find a pattern, or stress-testing a media strategy against multiple market scenarios, might. Using an LRM for simple tasks wastes cost. Using a standard LLM for tasks that require deep reasoning wastes accuracy.
It matters for client budget conversations. LRM inference is more expensive per query. Understanding why helps agencies justify the cost on appropriate use cases — and avoid recommending an expensive tool for tasks that don’t need it.
A strategist is using AI to evaluate three creative territory options for a new campaign. She runs the brief through a standard LLM first. It produces a summary that sounds plausible but misses a key tension in the brand positioning. She then runs the same prompt through an LRM with reasoning enabled. The model works through the brief more slowly, identifies the positioning conflict, and surfaces how each creative territory handles it differently. The output takes longer to arrive but becomes genuinely useful strategic input rather than surface-level commentary. The LRM is the right tool for this task.
The automations and agents module of the workshop teaches you how to build AI workflows that compress the busywork without taking the craft out of the studio.