AI Models
What the hell is the difference between Opus, Sonnet, GPT, and Gemini? And which one should you actually use?
1 How models work
What an AI Model Is, in Plain English: Parameters and Weights
What an AI language model is, in plain English: it guesses the next word, and the parameters (weights) are its settings - billions of them. Bigger means smarter, but pricier.
Tokens and the Context Window in Plain English
Tokens and the context window explained in plain English: why the model forgets the start of a conversation, why non-English text costs more tokens, and how you pay for input and output.
2 Who’s who
AI Model Families: Claude, GPT, Gemini, and the Open Ones
Language model families in plain English: what makes Claude, GPT, Gemini, and the open Llama, Qwen, Mistral, DeepSeek strong. Closed vs. open - what to pick for the job.
AI model price and speed: cheap vs. expensive
Cheap or expensive AI model - which one do you pick? We break down paying per token (input and output), why the answer costs more than the question, and how to split up tasks so your API bill drops by a wide margin.
How to Read AI Model Benchmarks and Not Fall for the Marketing
How to read AI model benchmarks and leaderboards: why the numbers lie (teaching to the test, cherry-picked metrics, test leaking into training) and how to compare models honestly on your own tasks.
3 Choosing for the job
How to Pick the Right AI Model for the Job: A Guide to Reasoning, Cost, and Context
How to pick the right AI model for the job: reasoning models for code and logic, cheap ones for classification, big context for documents, local ones for private data. Take the smallest model that does the job.
Multimodal Models: Screenshot to Code, Voice, Images, Video
Multimodality in plain English: a model that understands text, images, sound, and video and can create them too. Screenshot to code, voice instead of a keyboard, image generation - for vibecoders.