Instruct model
An instruct model is a base language model that has been further trained on instructions paired with good answers, and often on human feedback, so it follows requests instead of just continuing text.
Also called: instruction-tuned model, -Instruct, chat model
By Divya Sharma · Checked Nov 15, 2026 · How we review
When a model name ends in -Instruct, -it or -chat, it is usually the instruction-tuned version of a base model from the same release. The base model only continues text. The instruct model has learned to read a request and answer it.
How it is made
Starting from a pretrained model, the developer runs supervised fine-tuning on examples of instructions and the answers they want. Google's FLAN paper showed that this instruction tuning made models much better at tasks they had not seen. OpenAI's InstructGPT work then added a step where people ranked model outputs and the model was tuned further with RLHF. Newer models often use DPO or similar methods instead. Instruct models also learn a chat template: special tokens that mark system, user and assistant turns.
Base vs instruct
For most apps, use the instruct model. It follows a system prompt, answers questions, uses tools and refuses some requests. Pick the base model when you plan to do your own fine-tuning from a neutral start, or when you want plain text continuation, such as autocomplete.
Things to watch
Use the chat template the model was trained with. Many local runners apply it for you, but a wrong template is a common reason an open instruct model gives odd answers. Instruction tuning also changes behaviour in ways you may not want, such as long answers or refusals, and it does not make a model more truthful by itself.
Related terms
- Base model
- A base model is a language model straight out of pretraining, before instruction tuning or human-feedback training. It continues whatever text it is given rather than following instructions or holding a conversation.
- Supervised fine-tuning
- Supervised fine-tuning (SFT) trains a pretrained model further on example prompts paired with the responses you want. It is the usual first step that turns a base model into one that follows instructions.
- RLHF
- RLHF (reinforcement learning from human feedback) fine-tunes a language model using human rankings of its answers: a reward model learns those preferences, then reinforcement learning pushes the model toward answers that score well.
- System prompt
- A system prompt is the block of instructions an application sends to a language model before the user's messages. It sets the model's role, rules, tone and output format for the whole conversation.
- Pretraining
- Pretraining is the first and largest stage of training a model, where it learns general patterns from a very large unlabelled dataset, usually by predicting the next token, before any task-specific tuning.