gensai.aiA reasoning model is a language model trained, notably through reinforcement learning, to work through intermediate thinking before answering: it breaks the problem down, explores options and checks its result. More reliable in maths, code and analysis, it is also slower and more expensive. OpenAI o1 (2024) and DeepSeek-R1 (2025) led the way.
In practice at Gensai
Gensai reserves reasoning models for tasks that justify them, such as analysing complex files, and leaves simple answers to faster models.
Same theme
LLM (Large Language Model)SLM (Small Language Model)DLLM (Diffusion Large Language Model)Hallucination (AI)Prompt chainingPre-trained modelFoundation modelGenerative AITransformerTokenContext windowPromptPrompt engineeringContext (AI)Memory (AI)Fine-tuningOpen-weightQuantizationDistillationLoRA (Low-Rank Adaptation)Mixture of Experts (MoE)Temperature (AI)