mediumAI Engineering

What is the difference between a base model, an instruction-tuned model and a chat model?

Q: What is the difference between a base model, an instruction-tuned model and a chat model?

**TL;DR.** A **base model** is the raw next-token predictor — it completes text, it does not answer you. **Instruction tuning** teaches it to follow requests; **chat models** add multi-turn structure plus **preference training** (RLHF/RLAIF) for helpfulness and safety. Production APIs serve chat models. base model completes text instruction-tuned follows re

1.1k views

Understand the problem

Pretraining vs post-training: why raw next-token predictors need alignment before they are useful assistants.

base-modelinstruction-tuningrlhfpost-training

Attempt it yourself

Sketch your approach before reading the solution — that's what interviews test.

Stuck? AI Nudge Available

Get a conceptual hint to guide your logic without spoiling the final implementation.

Study the solution

The solution is waiting

Give it an honest attempt first — then compare your thinking with the full walkthrough.

Read the code

Same input, different behaviour

# base model — plausible continuation, not an answer
IN:  "What is the capital of France?"
OUT: " What is the capital of Spain? What is..."   # quiz-sheet pattern

# chat model — role-structured, answers the request
IN:  [system] You are a concise assistant.
     [user]   What is the capital of France?
OUT: "Paris."

Join the discussion

Discussion (0)

No responses yet. Be the first to share what you think.

How does streaming work in LLM APIs and why does it matter for UX?

How do you get reliable structured JSON output from an LLM?