Models & Providers Entry

Claude, explained

Reviewed October 2026

TL;DR: Claude is Anthropic's family of large language models. The weights are closed - you reach Claude through Anthropic's API, through major cloud platforms, or inside Anthropic's own apps and coding tools. The family's reputation rests on long-context work, careful instruction following, and strong performance as the engine behind coding agents.

What it is and how you use it

Claude is made by Anthropic, an AI company founded in 2021 by former OpenAI researchers with an explicit focus on AI safety research. Rather than a single model, Claude is a family released in named tiers - Haiku, Sonnet, Opus, and Fable - where the larger tiers maximize capability and the smaller ones trade some of it away for lower latency and cost. The tiers are versioned independently rather than refreshed together, so the durable way to think about Claude is as a capability ladder, not a list of version numbers.

The family is best known for a few things. It handles a large context window well, which suits reading long documents and whole codebases. It follows detailed instructions and system prompts closely, which matters when a model runs unattended inside a pipeline. And it is a common engine for AI coding agents and other agentic products, where reliable tool use over many steps counts for more than one-shot cleverness. Anthropic also trains the family with a technique it calls Constitutional AI, steering behavior with written principles.

Access is entirely hosted: Claude's weights are closed, so there is nothing to download and no local install. Developers call the first-party API directly or reach the same models through a cloud account, which matters for teams whose billing, compliance, and networking already live in one cloud. Those cloud routes split two ways: on Amazon Bedrock and Google Cloud's Agent Platform (the service formerly called Vertex AI) the cloud provider operates the inference stack and sets its own model retirement dates, while on Microsoft Foundry and Claude Platform on AWS Anthropic operates the models and the cloud supplies authentication, access control, and billing. Non-developers meet the family through the Claude apps and through Claude Code, which runs in the terminal, in an IDE, in a desktop app, and in the browser. If you need weights you can run yourself, look at an open family such as Llama instead.

In practice, teams reach for Claude when the work involves long documents, sustained multi-step agent runs, or code - and when they are comfortable with a hosted, per-token API rather than infrastructure they operate. It competes head-to-head with GPT and Gemini at the frontier tier, so the deciding factors are usually fit on your specific tasks, existing cloud commitments, and price, not any absolute ranking.

Where it sits in the AI stack

Claude occupies the model layer: your application composes a prompt, sends it over an API, and Anthropic or a cloud partner runs the inference. You never operate the model itself:

Key tools and implementations

  • Anthropic API

    The first-party HTTP API - the most direct path, and the first place new capabilities land.

  • Cloud platforms

    The same models through Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry, and Claude Platform on AWS - billed inside your existing cloud account, and operated by the cloud provider on the first two, by Anthropic on the last two.

  • Official SDKs

    Client libraries for Python, TypeScript, and other languages that wrap the API with streaming and tool-use helpers.

  • Claude apps and Claude Code

    Anthropic's own chat apps and its coding agent, which runs in the terminal, in an IDE, in a desktop app, and in the browser - the no-integration way to put the family to work.

Related entries