🟠 Claude Mastery

• The Claude Family: Opus, Sonnet, and Haiku (7 min) — ⚡ Anthropic organizes Claude into three tiers, each balancing intelligence, latency, and cost differently. As of mid-20…
• XML Tags: Claude's Native Structure (7 min) — Claude's training data included vast amounts of XML-structured text, which means XML tags are a first-class signal in i…
• Extended Thinking: Claude's Reasoning Mode (8 min) — Extended thinking is a mode available in Claude's API (claude-3-7-sonnet and later models) where the model performs an …
• Prompt Caching: Slash Cost and Latency with Reusable Context (8 min) — Prompt caching lets you mark a large, stable portion of your prompt — such as a system prompt, a long document, or a se…
• Tool Use & Agents 101 with Claude (8 min) — Claude supports structured tool use (also called function calling) where you define one or more tools in your API reque…
• Vision & Document Understanding (8 min) — Claude's vision capability lets you pass images (JPEG, PNG, GIF, WebP) and documents (PDF) directly into the conversati…
• Production Patterns with the Messages API (8 min) — The Anthropic Messages API is the primary interface for deploying Claude in production. It accepts a structured request…

The Claude Family: Opus, Sonnet, and Haiku

⚡ Anthropic organizes Claude into three tiers, each balancing intelligence, latency, and cost differently. As of mid-2026, the lineup is Haiku (fastest, cheapest), Sonnet (balanced default), and Opus (most capable, highest cost). Haiku is optimized for high-volume, low-complexit…

XML Tags: Claude's Native Structure

Claude's training data included vast amounts of XML-structured text, which means XML tags are a first-class signal in its attention mechanism — not a trick, but a genuine parsing aid. When you wrap distinct parts of a prompt in named tags, you reduce the chance that Claude will …

Extended Thinking: Claude's Reasoning Mode

Extended thinking is a mode available in Claude's API (claude-3-7-sonnet and later models) where the model performs an explicit internal reasoning pass before producing its final response. You enable it by setting `thinking: { type: 'enabled', budget_tokens: N }` in your API req…

Prompt Caching: Slash Cost and Latency with Reusable Context

Prompt caching lets you mark a large, stable portion of your prompt — such as a system prompt, a long document, or a set of few-shot examples — so the model stores its internal computation (the KV cache) after processing it the first time. Subsequent requests that include the sa…

Tool Use & Agents 101 with Claude

Claude supports structured tool use (also called function calling) where you define one or more tools in your API request. Each tool has a name, a description, and a JSON Schema that describes its parameters. When Claude decides a tool is needed, it returns a specially structure…

Vision & Document Understanding

Claude's vision capability lets you pass images (JPEG, PNG, GIF, WebP) and documents (PDF) directly into the conversation alongside your text prompt. Claude then reasons over both the visual content and your instructions together — no separate OCR step needed. Practical use case…

Production Patterns with the Messages API

The Anthropic Messages API is the primary interface for deploying Claude in production. It accepts a structured request with four key levers: **System prompt** (`system` field): A top-level instruction block sent before the conversation. Unlike user turns, it sets persistent per…