Edition and Model Snapshot

This is the Spring 2026 edition · 26 April 2026

The book is updated once a year, in spring, against the frontier models in wide use that quarter. If a chapter cites a model name or a specific behaviour, it refers to the version listed here. Your results may differ on later or earlier model versions.

Models referenced in this edition

Provider	Model	Used in book for
Anthropic	Claude Opus 4.7	Default chat model and Claude Code agentic CLI
Anthropic	Claude Sonnet 4.6	Cheaper everyday chat, batch jobs over CSVs
Anthropic	Claude Haiku 4.5	Cheapest classifier; sentiment-analysis baseline
OpenAI	ChatGPT 5.5	Cross-comparison; default for some readers
Google	Gemini 2.x	Mentioned in passing for long-context document tasks
Mistral	Le Chat / Mistral	EU-hosted alternative; mentioned in passing

The principles in the chapters are model-agnostic. The numbers, prompts, and screenshots are not — they were captured against the models above. When you read the book a year from now, expect the prompts to still work but the screenshots to look stale.

What changes year to year

Chapter 3 — Which AI Model. Re-shot every spring. Read this chapter on the website rather than from a print copy.
Chapter 16 — Data to Report. The end-to-end transcript is re-run on the current models so the comparison between vibe-report and directed-report stays honest.
Chapter 28 — APIs Under the Hood. Pricing tiers, rate limits, and structured-output APIs move every quarter; this chapter is the most dated bit of any printed copy.
The Capstone. The expectations-extraction prompt in Session 2 is re-tuned each edition because the model that runs it gets cheaper and smarter.

Previous editions

This is the first book edition. Earlier course editions (Spring 2025, etc.) ran as the website only. Course archives are available in the Git history when tagged.

Why update once a year, not every model release

Two reasons. One: a textbook needs a stable referent — students cite chapters, instructors plan against syllabi, errata accumulate. Two: a yearly cadence forces choices about what is durable (the principles, the workflow shape, the case studies) versus what is transient (a specific UI screenshot, a particular prompt verbatim). The transient bits get refreshed; the durable bits stay.

If a model release between editions changes the right answer to a chapter, an erratum will be filed in the repo’s issue tracker. Subscribe there.

--- title: "Edition and Model Snapshot" --- # Edition and Model Snapshot {.unnumbered} ::: {.callout-important} ## This is the **Spring 2026** edition · 26 April 2026 The book is updated once a year, in spring, against the frontier models in wide use that quarter. If a chapter cites a model name or a specific behaviour, it refers to the version listed here. Your results may differ on later or earlier model versions. ::: ## Models referenced in this edition | Provider | Model | Used in book for | |------------|------------------------|-----------------------------------------------------------| | Anthropic | **Claude Opus 4.7** | Default chat model and Claude Code agentic CLI | | Anthropic | **Claude Sonnet 4.6** | Cheaper everyday chat, batch jobs over CSVs | | Anthropic | **Claude Haiku 4.5** | Cheapest classifier; sentiment-analysis baseline | | OpenAI | **ChatGPT 5.5** | Cross-comparison; default for some readers | | Google | **Gemini 2.x** | Mentioned in passing for long-context document tasks | | Mistral | **Le Chat / Mistral** | EU-hosted alternative; mentioned in passing | The principles in the chapters are model-agnostic. The numbers, prompts, and screenshots are not — they were captured against the models above. When you read the book a year from now, expect the prompts to still work but the screenshots to look stale. ## What changes year to year - **Chapter 3 — Which AI Model.** Re-shot every spring. Read this chapter on the website rather than from a print copy. - **Chapter 16 — Data to Report.** The end-to-end transcript is re-run on the current models so the comparison between vibe-report and directed-report stays honest. - **Chapter 28 — APIs Under the Hood.** Pricing tiers, rate limits, and structured-output APIs move every quarter; this chapter is the most dated bit of any printed copy. - **The Capstone.** The expectations-extraction prompt in Session 2 is re-tuned each edition because the model that runs it gets cheaper and smarter. ## Previous editions This is the first book edition. Earlier *course* editions (Spring 2025, etc.) ran as the website only. Course archives are available in the [Git history](https://github.com/gabors-data-analysis/da-w-ai/tags) when tagged. ## Why update once a year, not every model release Two reasons. One: a textbook needs a stable referent — students cite chapters, instructors plan against syllabi, errata accumulate. Two: a yearly cadence forces choices about what is *durable* (the principles, the workflow shape, the case studies) versus what is *transient* (a specific UI screenshot, a particular prompt verbatim). The transient bits get refreshed; the durable bits stay. If a model release between editions changes the right answer to a chapter, an erratum will be filed in the repo's issue tracker. Subscribe there.