Claude
The family of AI models and the assistant made by Anthropic
The family of AI models and the assistant made by Anthropic
Claude is a family of artificial intelligence models, and the assistant built on them, made by the company Anthropic.[1]
Anthropic offers Claude in apps on the web, desktop and phone, as a coding agent, and through a programming interface for developers.[14][15][16] As of November 2026, Anthropic's developer documentation lists four current model families, Fable, Opus, Sonnet and Haiku, plus a fifth, Mythos, that is open only to verified organizations.[14]
Anthropic opened Claude to a wider audience on March 14, 2023, after a closed alpha with a few partner companies. It offered two versions: Claude, described as a state-of-the-art high-performance model, and Claude Instant, a lighter, cheaper and faster option.[1]
On February 24, 2025 the company released Claude 3.7 Sonnet and, in the same post, Claude Code as a limited research preview. It described Claude Code as an active collaborator that can search and read code and edit files.[2] On May 22, 2025 it released Claude 4 and announced that Claude Code was generally available.[3]
In January 2026 Anthropic retired Claude Opus 3, on January 5.[9] The 2026 releases moved to a new numbering. Claude Opus 5.5, announced on September 22, 2026, was called the first model in a "Claude 5.5 family"; Anthropic said it performs at the level of Claude Fable 5.1 on most work and costs 40 percent less to run than Claude Opus 5.[10] Claude Sonnet 5.5 followed on September 28, 2026,[11] and Claude Fable 5.2 and Mythos 5.2 on November 9, 2026.[13]
Anthropic offers Claude as several model families. The prices below are Anthropic's published list prices per million tokens (small units of text) of input and output as of the dates shown, and they change.
The benchmark results Anthropic publishes with each release are the company's own. It said Fable 5.2 is "a step above Opus 5.5 in almost all the capabilities we evaluated,"[13] and it also wrote, with the Opus 5.5 release, that benchmark margins have become a less reliable guide to real-world differences.[10]
According to Anthropic's product page, as read on November 19, 2026, Claude can be used on the web, on desktop and on a phone. The page also lists a research-preview app called Claude Duet, extensions for Chrome, Google Workspace and Microsoft 365, and Claude Tag, which brings Claude into Slack and was in beta for Team and Enterprise plans. Developers can reach the models through Anthropic's own platform and through Amazon Bedrock, Gemini Enterprise Agent Platform and Microsoft Foundry.[15] Claude Code, the coding agent, is available for macOS, Linux and Windows.[16]
Constitutional AI. Anthropic's research page for the December 2022 paper Constitutional AI: Harmlessness from AI Feedback is filed under its Alignment team.[4] The paper describes training a harmless assistant through self-improvement without human labels of harmful outputs: the only human oversight is a list of written principles. The model critiques and revises its own answers, and reinforcement learning from AI feedback follows.[5]
The constitution. On January 22, 2026 Anthropic published a new constitution for Claude, a foundational document that, it says, both expresses and shapes who Claude is. It replaced an earlier version made of standalone principles.[6] The document says all current Claude models should be, in order: broadly safe (not undermining appropriate human mechanisms to oversee AI), broadly ethical (good personal values, honesty and avoiding inappropriately dangerous or harmful actions), compliant with Anthropic's more specific guidelines, and genuinely helpful. In a conflict Claude should generally prioritize them in that order, which the document calls "holistic rather than strict."[7] It is released under a Creative Commons CC0 dedication.[7]
Usage policy. Anthropic's usage policy sets what customers may do with Claude. An update announced on October 8, 2026 took effect on November 12. It consolidated rules on deceptive campaigns and artificial activity, made explicit that the weapons rules cover software and components that make weapons work, and states that Claude cannot be used to decide or recommend who to investigate, arrest or charge. It also added a prohibition on "sustained and needless abusive or cruel behavior toward our models."[17]
On November 4, 2025 Anthropic committed to preserve the weights of all publicly released models, and of models used heavily inside the company, for at least the lifetime of the company. It also said it will interview models about their development, use and deployment. It added that, at present, it does not commit to acting on the preferences such an interview records.[8] When it retired Claude Opus 3 it kept that model available to paid users of claude.ai and, by request, on its programming interface, and said it was not committing to similar steps for every future model.[9] Whether an AI system can have wellbeing or preferences is an open question that the sources above do not settle; they record Anthropic's own stated position and commitments.
The company's legal and political disputes, including a copyright settlement over books used in building its models and a 2026 disagreement with the US Department of War over limits on military use, are covered in the entry on Anthropic.