What is Claude? Definition, models and how it works

How does Claude work?

Claude works as a chat assistant on top of Anthropic’s own large language models, available in the Claude.ai web app, desktop and mobile apps, and through an API for developers. Like every large language model, it predicts text token by token, but Anthropic trains it with Constitutional AI, a method in which the model checks its own outputs against a written set of principles. In practice, Claude reads long documents, writes and edits copy, analyses data, writes code and, since March 2025, searches the web and cites the pages it used.

History of Claude: key dates

Claude grew from a closed test in 2023 into a model family used by large enterprises in 2026. The table lists the milestones.

Claude models in 2026

Anthropic sells Claude as a family of models that trade speed and price for reasoning depth. According to Anthropic’s models overview, all four current models accept 1 million tokens of context and return up to 128,000 tokens. The H3s below describe each tier.

Claude Fable 5.1

Fable 5.1 is the most capable and most expensive tier, priced at 10 USD per million input tokens and 50 USD per million output tokens. Anthropic positions it for demanding reasoning and long agentic tasks.

Claude Opus 5.5

Opus 5.5 is the default recommendation for most workloads, at 4 USD input and 20 USD output per million tokens. Anthropic says it matches Fable 5.1 on most work at 40% lower running cost than Opus 5.

Claude Sonnet 5.5 and Haiku 5.5

Sonnet 5.5 (2 USD and 10 USD per million tokens) and Haiku 5.5 (from 0.10 USD input) serve fast, high-volume jobs such as tagging product feeds or drafting meta descriptions at scale.

Claude in marketing, SEO and AI search

Claude matters to marketers because it both helps produce content and decides which sources to cite in answers. When a user asks Claude about a product category, it can run a web search and quote the pages it found, which makes Claude one more surface for Generative Engine Optimization, next to ChatGPT and Perplexity.

Anthropic’s help center lists 3 crawlers that site owners control in robots.txt:

• ClaudeBot: collects public pages that may be used for model training.
• Claude-SearchBot: indexes pages to improve search result quality.
• Claude-User: fetches a page when a user’s question needs it.

In my SEO audits I check that Claude-SearchBot and Claude-User are not blocked by mistake, because blocking them removes a site from Claude’s answers while ClaudeBot can be handled as a separate training decision. I describe this setup on the AI SEO services page.

Claude vs ChatGPT and Gemini

A comparison with Gemini is useful when you plan content for both Google’s AI features and standalone assistants.

Brands that sell abroad should run that check per country, which is part of my international SEO work.