The History of Claude

Claude is Anthropic's family of large language models, and its history is shorter but no less important to AI coding: Anthropic shipped Claude Code, the agentic harness that defined the category.

Founding and Constitutional AI

Anthropic was founded in 2021 by Dario and Daniela Amodei and other former OpenAI researchers, on the thesis that AI safety should be the product, not an afterthought. The company's signature research contribution is Constitutional AI, described in a 2022 paper: instead of relying only on human preference ratings, the model is trained against a written constitution of principles (rooted in the UN Declaration of Human Rights, among other sources), has its own outputs critiqued against those principles, and is then fine-tuned to prefer responses that comply. The goal was a model that is helpful, honest, and harmless by construction.

Claude 1 through Claude 4

• Claude 1 (2023) shipped as a chat assistant with strong writing and a safety posture that made it conservative to the point of frustrating.
• Claude 2 (2023) opened the API more broadly and improved coding and reasoning.
• Claude 3 (2024) introduced the Opus, Sonnet, and Haiku tiering that has persisted, and was the first Claude that genuinely competed at the frontier.
• Claude 3.5 Sonnet became the default model for a large share of coding tools, including many early agent harnesses, because of its blend of speed and code quality.
• Claude 4 (2025) added extended thinking and consolidated the coding-first reputation. Its successors, Opus 4.5 through 4.8, steadily pushed context windows to 1M tokens and output limits to 128K.

Claude Code

In early 2025 Anthropic released Claude Code, a terminal agent that reads a codebase, plans, edits files, runs commands, and verifies its own work. It popularized several ideas that are now standard across harnesses: plan mode, project instruction files (CLAUDE.md), checkpoints for rolling back bad runs, subagents, hooks, skills, and MCP connections. For the purposes of this guide, Claude Code is the reference implementation of the harness layer, and most of the fundamentals in this course were written with it in mind.

Claude in 2026: Fable 5 and Mythos 5

Two new names arrived in June 2026:

• Claude Fable 5 is Anthropic's most capable broadly available model, positioned for long-running work with additional safeguards.
• Claude Mythos 5 is the same underlying model with some cybersecurity safeguards lifted for vetted Project Glasswing partners. It is deliberately not self-serve.

The pattern echoes OpenAI's tiering: a flagship for everyone, and a gated variant for a sensitive domain. For interview prep, the practical takeaway is that "Claude" is a family, and knowing which member you're using matters for reasoning about behavior.

Sources and further reading

• Anthropic: Constitutional AI
• Anthropic: Claude Fable 5 and Claude Mythos 5
• Anthropic: Project Glasswing and Mythos access