What Anthropic actually does

Claude comes in tiers that trade capability against speed and price. Haiku is the small, fast and cheap model for high-volume tasks. Sonnet is the workhorse most applications run on. Opus is the top general model, and Fable, a newer flagship tier priced above Opus, is aimed at the hardest and longest-running agent work. Customers choose a tier per task, and many use several.

Anthropic ships new generations quickly and tends to cut prices as it does. Claude Opus 5.5, released on September 22, 2026, performs at roughly the level of Fable 5.1 on most work, and Anthropic says it costs about 40% less than Opus 5 on typical workloads. Claude Sonnet 5.5 followed on September 28. Cheaper capability lets customers hand Claude more of their work; the bet is that the extra volume more than makes up for the lower price per task.

One model is deliberately not for sale to everyone. Claude Mythos, which Anthropic judged too capable at finding software vulnerabilities for an open release, is offered only to vetted security and critical-infrastructure organizations under Project Glasswing, launched in April 2026. Its early partners included Amazon, Apple, Google, Microsoft and Nvidia, and the EU's cybersecurity agency ENISA was given access in September 2026.

The Claude API is the core of the business. Developers send text, code, images or documents and pay for the tokens Claude reads and writes. It is how software companies build Claude into their own products, and how enterprises build their own agents on top of it.

Claude is also sold through Amazon Bedrock, Google Cloud's Vertex AI and Microsoft Foundry, which Anthropic says makes it the only frontier model available on all three of the largest clouds. Microsoft added it in November 2025, alongside a $30 billion Azure compute commitment from Anthropic. Selling through the clouds lets large companies buy Claude with the cloud budgets and contracts they already have. Microsoft also builds Claude into its own Copilot products, including GitHub Copilot and Microsoft 365 Copilot.

Anthropic also set the standard most AI agents use to connect to other software. It introduced the Model Context Protocol (MCP) in November 2024 as an open way for AI applications to reach tools and data, and in December 2025 handed it to the Agentic AI Foundation, part of the Linux Foundation, which it co-founded with Block and OpenAI. ChatGPT, Gemini, Microsoft Copilot and most coding tools now support MCP, and every integration a business builds for it also works with Claude.

Claude Code is an AI coding agent that works in a developer's terminal, editor or browser. It reads a codebase, edits files, runs tests and carries out multi-step programming tasks with little supervision. Anthropic released it as a preview in February 2025 and made it generally available in May 2025.

It became Anthropic's fastest-growing product. Six months after general availability it passed $1 billion in run-rate revenue, and by February 2026 it was above $2.5 billion, more than double its level at the start of that year. By then enterprises accounted for over half of Claude Code revenue, with customers such as Netflix, Spotify, KPMG, L'Oréal and Salesforce.

Coding matters to Anthropic well beyond Claude Code itself. Software development was the first job where AI agents reliably did hours of work at a time, and coding tools are among the largest buyers of the Claude API. That also makes it a crowded market: Claude Code competes with OpenAI's Codex, Google's coding tools, GitHub Copilot and Cursor, and the last two are also major Claude customers.

Claude is also an assistant on the web, desktop and phone, with a free tier and paid Pro, Max, Team and Enterprise plans. Paid plans add higher usage limits, Claude Code and the newest models. Claude in Chrome, Excel and PowerPoint puts it inside the tools office workers already use.

The apps are moving from answering questions to finishing whole tasks. Cowork, a desktop agent that works with a user's own files, launched in January 2026. On September 16, 2026 Anthropic folded it into the main Claude app and added Claude Docs and Claude Slides in beta, a direct challenge to Microsoft Office and Google Workspace. Claude Design, from Anthropic's Labs team, turns a description into designs, prototypes and slides.

Consumer use jumped in early 2026. Sign-ups surged in March, after Anthropic's public dispute with the Pentagon over how the military could use Claude, and kept growing in the weeks after. Businesses still bring in most of the money, but the apps are where many of those business users first meet Claude.

On top of the general products, Anthropic packages Claude for specific industries, with connectors to the data sources and systems each one already uses: Claude for Financial Services (July 2025), Claude for Life Sciences (October 2025), Claude for Healthcare for HIPAA-regulated work (January 2026) and Claude Science, a workbench for researchers (July 2026). In September 2026 it added a tool for financial advisers that plugs into the analytics and risk systems large asset managers use.

These products are as much a sales strategy as a technical one. They give Anthropic a reason to sell directly to banks, drug makers and hospitals, and each named partnership becomes a reference for the next customer in that industry. Novo Nordisk, for example, signed on in September 2026 to use Claude to speed up drug development.

Large accounts are the engine. In early 2024 only about a dozen customers spent more than $1 million a year with Anthropic. By February 2026 there were more than 500, and by April more than 1,000. A growing share of revenue comes from companies that have built Claude into their core operations rather than companies trying it out.

Some of the largest customers are other software companies. AI coding tools such as Cursor and GitHub Copilot buy Claude through the API at very large scale, and reporting in 2025 put the two among Anthropic's biggest accounts. That concentration cuts both ways. Cursor merged into SpaceX in August 2026, which also owns xAI, a direct competitor, and rents Anthropic part of its computing capacity. GitHub belongs to Microsoft, an Anthropic investor that also works closely with OpenAI.

Consultancies and software platforms resell Claude and build it into their own services. Deloitte rolled it out to more than 470,000 people in October 2025, and Accenture, Cognizant, Salesforce and Snowflake have partnerships of their own. In September 2026 Accenture also became the first outside firm to embed evaluators inside Anthropic to test its models.

The business is increasingly international. Anthropic opened its first Asian office in Tokyo in 2025, has since grown across Asia and Europe, and runs Claude inside India for local customers through Amazon Bedrock.

An independent body sits above the board. The Long-Term Benefit Trust, set up in 2023, holds a special class of shares that carries no economic rights but lets it elect directors, and its trustees have no financial stake in Anthropic. Since April 2026, when the Trust appointed Novartis chief executive Vas Narasimhan, Trust-appointed directors have made up a majority of the board. Shareholders can still change the arrangement, but only with a supermajority vote.

Anthropic also binds itself with published rules. Its Responsible Scaling Policy sets out the safety measures that have to be in place as its models become more capable, and the company has turned down business over how Claude may be used. Its refusal to let the Pentagon use Claude for mass domestic surveillance or fully autonomous weapons is what started the dispute described under the risks below.

The company was founded in 2021 by seven former OpenAI researchers: Dario Amodei, its chief executive, Daniela Amodei, its president, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish and Chris Olah. Ahead of the IPO, Anthropic has proposed a share class that would give the seven co-founders together 50.1% of the votes regardless of their economic stake, The Information reported in September 2026.

For who owns the company, see the investors page; for where the listing stands, see the IPO page.