Andrew Goldie, who writes as crimede-coder, published Large Language Models for Mortals with a specific promise: every code snippet in the book ran on the day it was compiled. Six months later, he recompiled the entire book to find out how badly the code had rotted. The answer, for most of it, is that it barely rotted at all.
The book contains 264 Python code cells that call OpenAI, Anthropic, Google, and AWS. Goldie did not update any of the pinned library versions, keeping openai at 2.14.0, anthropic at 0.75.0, and google-genai at 1.55.0. Three models had been retired by their providers. Everything else ran without changes.
Three model retirements, three targeted fixes
Claude Opus 4, which powered the extended thinking and web fetch examples, was replaced with Opus 4.5. Claude 3 Haiku, used in Chapter 4 to compare a cheaper model against Haiku 4.5, no longer had a cheaper Anthropic counterpart to pair with, so that section now compares Haiku 4.5 to Sonnet 4.5 at three times the price. Goldie's conclusion held: on a test set of 20 narratives, the two models produced near-identical results, so paying more bought nothing.
Gemini 3 Pro preview returned an API error pointing to gemini-3.1-pro-preview as its replacement, which worked as a drop-in swap. The many examples using gpt-4o-mini, AWS Bedrock, and agent tooling all ran as written.
The right calls and the wrong ones
Goldie admitted he was unsure whether foundation model providers would continue supporting fine-tuning, given how rarely he encountered it in practice. Custom-trained GPT models have since been deprecated. The book skipped fine-tuning entirely, so there was no code to fix.
The core API patterns held up. Chat completions, the responses API, structured outputs, and RAG implementations using in-memory stores, disk-based stores, and hosted vector stores all remained unchanged in ways that matter for the book's material.
On the agent front, Goldie guessed the Python agent libraries would undergo major changes. Six months later, the landscape looked largely the same. Codex now functions like Claude Code, and the OpenAI agents API folded into Codex, but no other fundamental shifts emerged from the foundation model providers. Most of the development activity has gone into building isolated environments around agents rather than changing how they query model endpoints.
The advice to have coding agents make small, reviewable changes also proved too conservative. Agents can now handle substantially larger tasks. The underlying principle, that a strong test suite is what makes agent-driven changes safe, has only become more relevant.
What will force the next update
AWS Bedrock now supports API keys, which Goldie called much simpler than the IAM role and inference profile setup his book walks through. The IAM configuration is likely still needed for other Bedrock services like S3 Vectors, but calling a model got easier.
The bigger shift came from open source. Goldie mostly omitted open weight models in the original book, showing only DeepSeek on Bedrock and a small Qwen model running locally. He did not anticipate how competitive models from GLM, Kimi, and DeepSeek would become. US-based endpoints from Databricks and Baseten now serve them, which matters for Goldie's work because sending data to servers outside the US is often a non-starter. He called AWS Bedrock "clearly falling behind" in this area.
Temperature=0, used throughout the book, is increasingly unsupported by newer thinking models. Goldie flagged that as the change most likely to force updates before a year passes. He may also need to deprecate the section on using logprobs for confidence scoring in classification, or adapt it to run on open source models instead.
No new edition planned for now
Goldie does not intend to release a revised version in the near term. He expects to publish a new edition in 2027, with previous buyers receiving the updated epub for free. The foundational topics, API calls, structured outputs, RAG, and tool calling, remain stable enough that the current book holds up.
He also offered promo codes for readers who want to evaluate the book before a revision appears: SIXMONTHSEPUB takes $30 off the epub, and SIXPRINT takes $20 off the paperback. Goldie's first book, Data Science for Crime Analysis with Python, remains the starting point for readers newer to Python.