Day 3 Exercise: API Parameter Experiments

This is the Day 3 companion exercise. Complete Day 1 first. Part 1 covers the theory (“Understanding API Parameters”)—today you verify it with your own eyes.

Part 1 explained parameters in theory. But theory without practice is just noise. Today you run three experiments and see for yourself how parameters affect output.

Setup

Make sure your Day 1 project still works:

uv run python hello_api.py

If the AI replies, your environment is ready. All experiments below build on this code.

Experiment 1: Temperature from 0 to 1

Copy hello_api.py as experiment_temp.py and change it to:

import os
from dotenv import load_dotenv
from openai import OpenAI

load_dotenv()

api_key = os.environ.get("DEEPSEEK_API_KEY")
if not api_key:
    raise ValueError("DEEPSEEK_API_KEY not set. Check your .env file.")

client = OpenAI(
    api_key=api_key,
    base_url="https://api.deepseek.com"
)

question = "Write a short story about a cat (three sentences max)"

for temp in [0, 0.5, 1.0]:
    print(f"\n{'='*40}")
    print(f"temperature = {temp}")
    print('='*40)

    response = client.chat.completions.create(
        model="deepseek-v4-flash",
        messages=[{"role": "user", "content": question}],
        temperature=temp
    )

    print(response.choices[0].message.content)

Run it:

uv run python experiment_temp.py

What to observe:

• temperature=0: Nearly identical output every time. AI picks the highest-probability word. No randomness.
• temperature=0.5: Some variation, but the logic remains coherent.
• temperature=1.0: Noticeably different each run. Word choice, style, and story direction may vary.

Try it: Run the same code two or three times and compare the stability of temperature=0 vs temperature=1.0.

When to use which:

• Deterministic output (code, data extraction) → temperature=0
• Some variation but controlled (daily conversation, translation) → temperature=0.5-0.7
• Creative work (writing, brainstorming) → temperature=0.8-1.0

Experiment 2: max_tokens Truncation

Create experiment_tokens.py:

import os
from dotenv import load_dotenv
from openai import OpenAI

load_dotenv()

api_key = os.environ.get("DEEPSEEK_API_KEY")
if not api_key:
    raise ValueError("DEEPSEEK_API_KEY not set. Check your .env file.")

client = OpenAI(
    api_key=api_key,
    base_url="https://api.deepseek.com"
)

question = "Explain what machine learning is, with three real-life examples"

for limit in [50, 200, 1000]:
    print(f"\n{'='*40}")
    print(f"max_tokens = {limit}")
    print('='*40)

    response = client.chat.completions.create(
        model="deepseek-v4-flash",
        messages=[{"role": "user", "content": question}],
        max_tokens=limit
    )

    print(response.choices[0].message.content)
    print(f"\n(actual output tokens: {response.usage.completion_tokens})")

Run it:

uv run python experiment_tokens.py

What to observe:

• max_tokens=50: AI starts explaining but gets cut off mid-sentence.
• max_tokens=200: A short answer, but probably not enough room for all three examples.
• max_tokens=1000: Full answer with all three examples expanded.

Note: max_tokens is a ceiling, not a target. AI won’t pad output to fill the limit. A 100-token answer won’t grow to 1000 tokens just because you raised the ceiling.

Experiment 3: Calculate Your Actual Cost

Add this to the end of experiment 2:

    # Cost estimate for the last request
    usage = response.usage
    input_tokens = usage.prompt_tokens
    output_tokens = usage.completion_tokens

    # DeepSeek V4-Flash pricing (May 2026, subject to change)
    input_price = 0.14 / 1_000_000   # $/token
    output_price = 0.28 / 1_000_000

    cost = (input_tokens * input_price) + (output_tokens * output_price)
    print(f"\nRequest cost: ${cost:.6f} (approx ¥{cost * 7.2:.4f})")
    print(f"  Input: {input_tokens} tokens, Output: {output_tokens} tokens")

Build Your Intuition

• A simple Q&A (100 input + 200 output tokens) costs about $0.00008 (roughly ¥0.0006)
• $1.50 (roughly ¥10) can fund roughly 15,000–20,000 such requests
• Long texts are where costs add up: summarizing a 5,000-character article may consume 7,500 input tokens

What You Did Today

• Ran the temperature experiment and saw the difference between 0 and 1.0
• Ran the max_tokens experiment and saw truncation in action
• Calculated the actual cost of one API request

Tip: Not sure how to tune parameters? Describe the output you want in plain language and ask AI: “I want more stable/creative/shorter output, what should I set temperature and max_tokens to?”

Next up: Day 4 is Part 2 of the main tutorial—moving from single calls to batch processing. You’ll learn to have the API process 100 files automatically.