Definition
Claude is Anthropic's family of language models. Its named lines include Opus, Sonnet, Haiku, Fable, and Mythos. A line contains successive releases; a name such as Claude Sonnet 5.5 identifies one release within that family.
Anthropic also uses Claude for its assistant product. A conversation in Claude.ai and a direct call through the Claude API can involve different instructions, tools, memory, and access rules. Comparing those systems requires more detail than the shared brand.
Origin and attribution
Anthropic introduced Claude publicly on March 14, 2023, following a closed alpha with partners. It offered Claude and the lighter Claude Instant. On March 4, 2024, it introduced the Claude 3 family and the Opus, Sonnet, and Haiku names. Fable and Mythos are later additions to the product range.
The attribution belongs to Anthropic's model-development teams. The announcements do not identify one person as the inventor of the family or its names.
Scope and limits
Claude names a vendor's products, not a neural architecture or a fixed capability threshold. Current general-availability models accept text and image inputs and produce text. Features such as supported thinking modes, tool interfaces, context limits, and safeguards vary by release and serving platform.
Anthropic distinguishes a fixed model snapshot from the serving infrastructure around it, which can change observed behavior. To assess a performance or safety claim, an evaluation needs the exact model identifier and the surrounding configuration.
Operational significance
Record the release, provider, effort setting, prompts, and tools when comparing results. Check the migration guide before switching releases, and repeat the tests that cover the application's failure costs. Keep authorization and verification in the application even when the chosen model performs well in the vendor's evaluations.
Distinguish it from nearby terms
- Opus, Sonnet, Haiku, Fable, and Mythos are lines or deployment names within Claude. They are not interchangeable releases.
- A large language model describes a broad technical class. Claude is one branded family in that class.
- An agent harness supplies the control loop, tools, and state around a model. Claude Code is an application built around Claude models, with its own execution behavior.
- Model routing selects which model serves a request. A product can use several Claude releases during one workflow.
Check your understanding
Two teams report that they tested "Claude," but one used Claude.ai and the other called a pinned Sonnet release through an API with custom tools. Which details would you need before comparing their results?