Learn about models and agents

How models are priced, what a context window really holds, and how to bring your own key.

Guides

How model pricing worksInput, output, cached reads, cached writes and reasoning tokens — the five numbers on an invoice, and which one usually surprises you. 11 Jun 2026Context windows, and why you never get the whole numberSystem prompts, tool definitions and reserved output all eat the window before your text does. 18 Jun 2026Bring your own key: what each provider asks forOne field for OpenAI, four for Azure, a service account for Vertex, keys or a role for Bedrock. 25 Jun 2026Choosing a model for a jobCapability, latency, price and context — in the order that actually matters for the thing you are building. 2 Jul 2026Reasoning models, and when they are worth the tokensWhat thinking before answering costs, and the classes of problem where it pays for itself. 9 Jul 2026Tool use and structured outputHow a model calls your code, why schemas beat prompting, and where it still goes wrong. 16 Jul 2026Prompt caching, and how to actually benefit from itCache writes cost more than normal input. Get the ordering wrong and caching makes you slower and poorer. 23 Jul 2026

Articles

The same model costs different amounts depending on where you run itOne model, four hosts, four identifiers, four prices — and the cheapest is not automatically the right one. 30 Jul 2026Running inference in Europe, honestlyWhich providers actually keep your requests in the EU, and which merely bill you from it. 6 Aug 2026

Everything we have written →