Claude (AI)

Template:Infobox software

Claude is a family of large language models developed by the AI safety company Anthropic. Named after Claude Shannon, the founder of information theory, Claude is positioned as a conversational AI assistant designed around Anthropic's research agenda on AI alignment and safety. Claude is accessible through the consumer chat interface at claude.ai, the Anthropic API, and through partnerships with cloud providers including Amazon Bedrock and Google Cloud Vertex AI.

History

Anthropic was founded in 2021 by former OpenAI researchers including siblings Dario and Daniela Amodei, along with several colleagues who left OpenAI over concerns about the direction of large-scale AI development. Claude was first released to a limited group of partners in March 2023, and to the general public via the claude.ai web interface later that year.

Subsequent model generations included Claude 2 (July 2023), the Claude 3 family — Haiku, Sonnet, and Opus — (March 2024), Claude 3.5 Sonnet (June 2024), Claude 3.7 Sonnet (early 2025), and the Claude 4 family beginning in 2025. As of 2026, the current generation includes Claude Opus 4.6, Claude Sonnet 4.6, and Claude Haiku 4.5, organised around a tiered "Haiku / Sonnet / Opus" naming scheme corresponding to increasing capability and cost.

Model family

Claude models are released in tiers, with each tier balancing intelligence, latency, and cost differently:

Haiku — the smallest and fastest tier, optimised for low-latency, high-volume tasks such as classification, extraction, and lightweight agentic loops.
Sonnet — the mid tier, intended as a general-purpose workhorse balancing capability and cost. Sonnet models have historically been the default for most coding and reasoning workloads.
Opus — the largest and most capable tier, intended for complex reasoning, long-horizon agentic tasks, and frontier research use cases.

All current Claude models are multimodal in input, accepting text and images, and emit text output. They support extended context windows; the Claude 4-series Opus models are offered with context windows up to one million tokens for select customers.

Constitutional AI

A defining feature of Claude's training methodology is Constitutional AI (CAI), an approach Anthropic introduced in a 2022 paper. Rather than relying solely on reinforcement learning from human feedback (RLHF) for harmlessness training, CAI uses a written set of principles — a "constitution" — to guide a model in critiquing and revising its own outputs. The constitution draws on sources such as the Universal Declaration of Human Rights and Apple's terms of service, and is designed to be transparent and auditable.

Constitutional AI is used alongside RLHF and, in later models, reinforcement learning from AI feedback (RLAIF), in which model-generated preferences supplement or partially replace human raters.

Claude Code

Claude Code is Anthropic's official command-line interface for software engineering tasks, released in 2025. It provides an interactive terminal agent built on the Claude models and the Claude Agent SDK, with tools for reading and editing files, running shell commands, and orchestrating subagents. Claude Code is also available as a desktop application, web application at claude.ai/code, and as IDE extensions for Visual Studio Code and JetBrains products.

Claude Code popularised several patterns now common in agentic coding tools, including persistent project instructions via Template:Code files, hook-based automation, and the Model Context Protocol (MCP) for connecting external tools and data sources.

Model Context Protocol

Template:Main In late 2024 Anthropic published the Model Context Protocol (MCP), an open standard for connecting AI assistants such as Claude to external data sources, tools, and services. MCP defines a JSON-RPC-based interface in which a client (the AI assistant) connects to one or more servers exposing tools, resources, and prompts. The protocol has been adopted by other vendors and is now widely used as a vendor-neutral plugin ecosystem for LLM-based agents.

Reception

Claude has been broadly well-received, particularly for tasks involving long-form writing, code generation, and tool use. Independent benchmark suites and user surveys throughout 2024 and 2025 frequently ranked Claude Sonnet and Opus models among the strongest available for software engineering, with Sonnet 3.5 and the later Sonnet 4 / Opus 4 generations becoming popular defaults in agentic coding tools. Critics and supporters alike have noted that Claude tends toward a more cautious refusal posture than some competitors, a deliberate consequence of Anthropic's safety-first positioning.

References

Bai, Y. et al. (2022). "Constitutional AI: Harmlessness from AI Feedback." arXiv:2212.08073.
Anthropic. "Introducing Claude." Anthropic blog, March 14, 2023.
Anthropic. "Introducing the Model Context Protocol." Anthropic blog, November 2024.
Anthropic. "The Claude 3 Model Family: Opus, Sonnet, Haiku." Anthropic, March 2024.

External links