AI guides

Our guides explain the concepts you need to read AI model data with confidence: how token prices work, what a context window really limits, when a reasoning model is worth its cost, and how to compare models without being misled by a single number. Each guide answers a practical question and points to the live data on this site, so you can go straight from understanding a concept to applying it. The full list of guides is at the end of this page.

Find the right guide for your question

"What will this cost me?"

Start with AI token pricing explained. It explains how APIs bill per token, why output usually costs more than input, and how cached input, batch processing and prompt-length tiers change the bill. How to reduce AI API costs builds on it with ways to spend less: caching repeated prompt parts, batching work that is not urgent, cutting output and reasoning tokens and routing simple tasks to smaller models. To put numbers on your own workload, use the AI cost calculator; to compare list prices directly, use AI pricing.

"Will my documents fit?"

Context windows explained covers how much text a model can take in one request, how the window differs from the maximum output, where long-context surcharges start and why a model can lose track of details long before its limit.

"Which model should I use?"

How to choose an AI model walks through the decision step by step: describe the task, rule out models by inputs and outputs, context window, license, deployment and data policy, then test quality and weigh latency against cost. How to compare AI models explains how to read benchmarks, prices and rankings side by side: why scores only compare within the same benchmark setting, what "self-reported" means, why a small score difference rarely settles a choice and how to normalize prices for your own mix of input and output. Our compare page and the model directory put that into practice.

"Do I need a reasoning model?"

Reasoning models explained covers the difference between models without extended reasoning, models where it can be switched on and models where it is always on, and how reasoning tokens add to cost and response time.

"Can it work with images, audio or video?"

Multimodal AI explained separates what a model can read from what it can produce, and explains how image generation is billed: per image or per image token, not per text token.

"How do I turn AI content into a publication?"

AI images, video, audio and 3D for a digital magazine follows a small team through one issue: which kind of model suits text, images, video, voice and 3D, which file formats to ask for, how the finished PDF becomes an interactive magazine, and what to check for labels, accessibility and cost.

"How do I get better answers?"

Prompt engineering basics covers prompt structure, examples, output formats and how to test changes. It also shows the cost side: input tokens are billed on every request, so a long system prompt is worth caching.

Numbers worth knowing

These figures come from the models and list prices in our catalog as of October 2026. They change as providers update their offers; AI pricing always shows the current values.

• Output costs more than input. Among text models that list both prices, output tokens cost a median of four times as much as input tokens. The ratio ranges from 1:1 to just over 8:1.
• Cached input is much cheaper. It is most often billed at a tenth of the normal input price; across the models that list it, the range is 2% to 25%.
• Batch processing usually halves the price. Of the text models with batch prices, 30 of 37 charge half the standard rate and the other seven take 20% off.
• Prices span a wide range. Base-tier input prices run from $0.035 to $10 per 1M tokens, output prices from $0.10 to $50. The cheapest and the most expensive models differ by a factor of more than 250 on input.
• Large context windows are common. Just over half of the text models with a documented context window accept 1 million tokens or more. Documented windows range from about 65,000 to 10 million tokens.
• Most models can reason. Of the 60 text models whose reasoning support is documented, 51 offer extended reasoning, either switchable or always on.

How the guides are written

Guides explain concepts and cite sources for their factual claims, mostly provider documentation and pricing pages. Where a guide shows prices or specs, it embeds live tables from our database where possible, so the figures update with the catalog; numbers written into the text carry an "as of" date. Each guide shows when it was published or last updated, and guides are flagged for review every 180 days. The methodology page explains where our data comes from and how we check it.