Thinking Model

Introduction

When you ask most AI models a question, they give a direct answer—sometimes correct, sometimes not. But what if we want the AI to reason like a human before answering?

This is where Chain-of-Thought (CoT) prompting comes in. It’s a way to make even a non-thinking model act like a thinking model by guiding it to show its reasoning process.

What is a Non-Thinking Model?

A non-thinking model is like a student who blurts out the final answer without showing their work.

Example: You ask, “What is 37 × 45?”
AI might reply directly: “1665”

✅ Quick, but ❌ no reasoning.

What is Chain-of-Thought?

Chain-of-Thought prompting tells the model:
👉 “Don’t just give the answer. Think step by step. Show your reasoning.”

So instead of blurting out:

It breaks the problem into smaller parts
Checks correctness at each step
Then delivers the final answer

Example: With and Without CoT

Without CoT:
User: “Solve 23 + 47 × 2”
AI: “117” (might be wrong due to missing order of operations)

With CoT:
AI:

Step 1: Follow BODMAS → multiply first.
Step 2: 47 × 2 = 94
Step 3: 23 + 94 = 117
Final Answer: 117 ✅

Here, the process is transparent and trustworthy.

How CoT Builds a Thinking Model

START Phase: The model identifies the problem.
THINK Phase: It explores possible solutions step by step.
EVALUATE Phase: It double-checks correctness.
OUTPUT Phase: It presents the final result.

This process transforms the model from a “black-box answer machine” into a “transparent problem solver.”

Code Example: Chain-of-Thought with Gemini

import "dotenv/config";
import { OpenAI } from "openai";

const client = new OpenAI({
  apiKey: process.env.GEMINI_API_KEY,
  baseURL: "https://generativelanguage.googleapis.com/v1beta/openai/",
});

const SYSTEM_PROMPT = `
You are an assistant who always reasons step by step using:
START → THINK → EVALUATE → OUTPUT
`;

async function main() {
  const response = await client.chat.completions.create({
    model: "gemini-2.5-flash",
    messages: [
      { role: "system", content: SYSTEM_PROMPT },
      { role: "user", content: "Solve 12 * 8 + 50" },
    ],
  });

  console.log(response.choices[0].message.content);
}

main();

This ensures the AI doesn’t just give the final number—it shows how it got there.

Why It Matters

Trust: You can see how the answer was derived.
Debugging: If AI goes wrong, you know where.
Better Accuracy: Step-by-step reasoning reduces silly mistakes.

Final Thoughts

Chain-of-Thought prompting is like teaching a student to show their work instead of guessing. By encouraging AI to break down problems step by step, we convert a non-thinking model into a thinking partner.

How Chain-of-Thought Turns a Non-Thinking AI into a Thinking Mode