1.2 Free Options & Zero-Cost Start Beginner Free

Prerequisites: 1.1 Your First AI Conversation

Why Do You Need It? (Problem)

In the previous section, we used OpenAI API to complete our first AI conversation. But you might immediately have questions:

• "API calls cost money, I'm still learning and don't want to incur costs"
• "My company's network has unstable access to OpenAI"
• "Company data is sensitive and cannot be sent to overseas cloud services"

Good news: there are multiple free or extremely low-cost options to solve these problems.

What Is It? (Concept)

LLM usage methods can be divided into three categories:

Option	Cost	Advantages	Suitable Scenarios
Google AI Studio	Generous free tier	Gemini models, Google quality	First choice for learning
OpenRouter Free Models	$0	One API for multiple free models	Want to try various models
DeepSeek	Very low (~1/10 of OpenAI)	Domestic model, strong Chinese capability	Budget-sensitive, Chinese scenarios
Ollama (Local)	$0	Data stays local, fully offline	Sensitive data, offline environments

Try It Yourself (Practice)

Option 1: Google AI Studio (Recommended for Beginners)

pip install google-genai

from google import genai

client = genai.Client(api_key="YOUR_GOOGLE_API_KEY")
response = client.models.generate_content(
    model="gemini-2.0-flash",
    contents="Explain what an API is in one sentence",
)
print(response.text)

Get API Key

Go to Google AI Studio to get it for free, no credit card required.

Option 2: OpenRouter Free Models

OpenRouter is a model aggregation platform offering multiple free models. The best part: it's compatible with OpenAI's API format—just change two lines of code:

from openai import OpenAI

client = OpenAI(
    base_url="https://openrouter.ai/api/v1",
    api_key="YOUR_OPENROUTER_API_KEY",
)
response = client.chat.completions.create(
    model="deepseek/deepseek-chat-v3-0324:free",
    messages=[{"role": "user", "content": "Explain what an API is in one sentence"}],
)
print(response.choices[0].message.content)

Get API Key

Go to OpenRouter to register, free models don't require top-up.

Option 3: DeepSeek (Extremely Low Cost)

DeepSeek is a high-quality domestic model with prices about 1/10 of OpenAI's, excellent Chinese capability:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.deepseek.com",
    api_key="YOUR_DEEPSEEK_API_KEY",
)
response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[{"role": "user", "content": "Explain what an API is in one sentence"}],
)
print(response.choices[0].message.content)

Price Comparison

• OpenAI GPT-4o-mini: $0.15 / million input tokens
• DeepSeek Chat: ~¥0.1 / million input tokens (less than 1/10 of OpenAI)

Option 4: Ollama Local Running (Completely Free)

Ollama lets you run open-source LLMs on your own computer, data completely stays local:

# 1. Install Ollama (macOS)
brew install ollama

# 2. Start and download model (first time requires download, ~4GB)
ollama run qwen2.5:7b

After installation, Ollama automatically provides a local API compatible with OpenAI format:

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ollama",
)
response = client.chat.completions.create(
    model="qwen2.5:7b",
    messages=[{"role": "user", "content": "Explain what an API is in one sentence"}],
)
print(response.choices[0].message.content)

Hardware Requirements

• 7B models: at least 8GB RAM
• 14B models: at least 16GB RAM
• 70B models: require 64GB+ RAM or professional GPU

Core Discovery: Unified API Format

You may have noticed: the code for all 4 options above is almost identical—all use the OpenAI client, just changing base_url and api_key.

This is not a coincidence. OpenAI's Chat Completions API format has become the de facto industry standard, with almost all platforms being compatible. This means you can write one codebase and switch model providers anytime.

import os
from openai import OpenAI

PROVIDERS = {
    "openai":    {"base_url": "https://api.openai.com/v1",       "model": "gpt-4o-mini"},
    "deepseek":  {"base_url": "https://api.deepseek.com",        "model": "deepseek-chat"},
    "openrouter":{"base_url": "https://openrouter.ai/api/v1",    "model": "deepseek/deepseek-chat-v3-0324:free"},
    "ollama":    {"base_url": "http://localhost:11434/v1",        "model": "qwen2.5:7b"},
}

provider = os.getenv("LLM_PROVIDER", "openrouter")
config = PROVIDERS[provider]

client = OpenAI(base_url=config["base_url"], api_key=os.getenv("LLM_API_KEY", "ollama"))
response = client.chat.completions.create(
    model=config["model"],
    messages=[{"role": "user", "content": "Explain what an API is in one sentence"}],
)
print(response.choices[0].message.content)

Run locally: jupyter notebook demos/01-first-experience/free_options.ipynb

Summary (Reflection)

• What Did This Solve: Can learn AI programming without spending money, multiple options for different needs
• What Remains Unsolved: So many models, big differences in capabilities, how to systematically understand them? → Chapter 2: AI Landscape & Model Platforms
• Key Points:
  1. Google AI Studio and OpenRouter free models are the best entry points for zero-cost learning
  2. Ollama local deployment suits sensitive data and offline scenarios
  3. OpenAI API format is the industry standard—learn once, use everywhere

---

Getting Started Checkpoint (Chapter 1 Part)

Complete the following tasks to confirm you've mastered Chapter 1 core knowledge:

• [ ] Complete AI conversations using at least 2 different options (e.g., OpenAI + Ollama, or Google AI Studio + OpenRouter)
• [ ] Implement a simple script that supports multi-turn conversation
• [ ] Can explain the roles of system/user/assistant three message types

---

Last Updated: 2026-02-20