Working with AI Reasoning Models
Published by O'Reilly Media, Inc.
How and when to use LLMs for thinking and reasoning
Course outcomes
- Differentiate between traditional LLMs and reasoning-focused models like OpenAI’s GPT-Thinking series, DeepSeek, Kimi, Claude Sonnet/Opus series and Gemini Pro series.
- Design effective prompts specifically for reasoning-focused LLMs
- Evaluate and select appropriate models based on metrics like reasoning quality, embedded knowledge, and context window requirements
- Develop systematic approaches for testing and improving reasoning model outputs
Course description
Join expert instructor Lucas Soares to explore the newest generation of large language models specifically designed for thinking and reasoning tasks. You’ll learn how these models differ from traditional LLMs, when to use them, and how to leverage their advanced capabilities effectively. Through hands-on exercises and real-world examples, using a mix of regular chat interfaces and Python code, you’ll master the techniques for prompting these models, evaluating their outputs, and integrating them into your applications.
What you’ll learn and how you can apply it
- Learn the fundamentals of reasoning-focused LLMs and their architectures
- Learn when and how to use models like GPT-5.4-Thinking, Claude Opus 4.6, Gemini 3.1 Pro
- Develop effective prompting strategies for reasoning tasks
- Implement best practices for model selection and evaluation
- Understand the trade-offs between different model types
This live event is for you because...
- You’re a developer, AI engineer, or data scientist who wants to build applications that require complex reasoning capabilities.
- You want to understand when to use reasoning-focused LLMs versus traditional models.
- You want to improve the quality and reliability of AI-powered solutions.
- You need to stay current with the latest developments in LLM technology.
Prerequisites
- A basic understanding of modern LLMs
- Experience with programming languages such as Python
- Familiarity with RESTful APIs and web services (helpful but not required)
Recommended follow-up:
- Take Using LLMs for Software Engineering (live online course with Chelsea Troy)
- Read Hands-On Large Language Models (book)
- Read An Illustrated Guide to AI Agents (early release book)
- Read Designing Large Language Model Applications (book)
Schedule
The time frames are only estimates and may vary according to how the class is progressing.
Introduction to Reasoning Models (50 minutes)
- Presentation: Introduction to Reasoning models - LLMs recap;defining reasoning; CoT; reasoning output structure; timeline of reasoning models; UIs and pricing; benchmarks; key metrics.
- Q&A
- Break
How Reasoning Models Are Trained (70 minutes)
- Presentation: How Reasoning models are trained - the problem; reward signals; R1-zero training curve; GRPO vs PPO; output length issues; DeepSeek R1; distillation.
- Hands-on: Spotting reasoning models in the wild (chat interfaces), testing local and closed source reasoning models using standard APIs (Ollama, OpenAI, Anthropic).
- Q&A
- Break
Prompting Reasoning Models: What's Different and What Works (60 minutes)
- Presentation: Prompting Reasoning Models - reasoning models as "senior coworkers" (give goals, not step-by-step procedures) versus traditional LLMs as "junior coworkers" (give explicit instructions).
- Hands-on exercise: Take a real task and write two prompts — one traditional style with step-by-step instructions, one reasoning-model style with goals and constraints; run both through a reasoning model API, compare output quality and token usage
- Q&A
- Break
Engineering Decisions: When to Reason, How Much, and Is It Worth It (60 minutes)
- Presentation: The usage/cost framework for employing reasoning models in your workflows
- Hands-on exercise: Build a contract clause risk analysis pipeline— run the same task at three effort levels, log token counts and dollar costs, identify the minimum effort level that meets a quality threshold
- Q&A
Your Instructor
Lucas Soares
Lucas Soares is a machine learning engineer who has worked at K1 Digital and Biometrid, where he developed computer vision and NLP models for applications such as document verification, OCR-based applications, and recommender systems. Lucas has also developed various ML models, including neural networks, Siamese networks, convolutional neural networks, LSTMs, and genetic algorithms.