What is Claude, and who is Anthropic?
Anthropic is an AI safety company founded in 2021 by Dario Amodei, Daniela Amodei, and several colleagues who previously worked at OpenAI. Their central argument is simple but significant: the labs building the most powerful AI should also be the most focused on making it safe. That belief shapes everything about how they build and release their models.
Claude is Anthropic's flagship AI — a large language model you can talk to, ask questions, have write or edit documents, help with research, write code, and much more. The name is a nod to Claude Shannon, the mathematician whose work on information theory laid the groundwork for modern computing and communication.
What Claude is known for, beyond raw capability, is the quality of its writing and the care it takes in how it responds. It tends to be thoughtful, precise, and less prone to making things up than some competing models. It's used by individuals, businesses, researchers, and developers across a wide range of industries.
The Past — Claude 3
Claude 3 launched in early 2024 and was a significant milestone for Anthropic. It introduced the three-tier naming system that Claude still uses today: Haiku, Sonnet, and Opus.
The names reflect a deliberate philosophy about how AI models should be offered. Not every task requires the most powerful model available — sometimes you need speed and low cost, sometimes you need balance, and sometimes you genuinely need maximum capability. The three tiers are designed for those three scenarios.
- Haiku — named after the compact Japanese poem form — is the fastest and lightest model. Built for tasks where speed and volume matter more than depth.
- Sonnet — a mid-length poem form — is the balanced tier. Capable enough for most professional work, faster and cheaper than Opus.
- Opus — the longest and most complex poem form — is the most capable model in the family, for demanding, multi-step work.
Claude 3 was widely praised and put Anthropic on the map as a serious competitor to OpenAI and Google. Before moving to Claude 4, Anthropic released several incremental updates — Claude 3.5 and Claude 3.7 — that improved on the originals without changing the fundamental structure. Claude 3 models have largely been superseded, but some remain in use.
The Present — Claude 4
Claude 4 is the current generation. The Haiku/Sonnet/Opus structure carries over, but the capabilities, pricing, and context windows have all changed significantly.
Haiku 4.5
Haiku 4.5 is the fastest and cheapest model in the Claude 4 family. It has a 200,000-token context window — large enough to process a long document or extended conversation, though smaller than the other Claude 4 models.
Its strengths are speed and efficiency. It's well-suited to tasks that need to happen quickly or at high volume: answering short questions, classifying content, summarizing documents, powering customer-facing chat features where response time matters.
Who this is for: Primarily developers building products that need to handle many requests cheaply and quickly. If you're a general user choosing between models on Claude.ai, Haiku is the right pick for simple, fast tasks — not for anything requiring deep reasoning or nuance.
Sonnet 4.6
Sonnet 4.6 is the model most people are using when they use Claude. It's Anthropic's recommended default for the vast majority of use cases, and it's the first Sonnet to outperform the previous generation's Opus on coding benchmarks — a meaningful jump in capability.
It has a 1-million-token context window. To put that in plain terms: you could paste in an extremely long document, an entire book, or a large codebase and Claude can work with all of it at once without losing track of what came earlier.
Best for: professional writing, research and summarization, coding assistance, document analysis, drafting emails and reports, and most everyday AI tasks.
Who this is for: Almost everyone. If you're a professional using Claude to help with your work — a teacher, marketer, nurse, analyst, or anyone else — Sonnet is the tier you'll use most of the time. It's capable enough to handle serious tasks and fast enough not to feel slow.
Opus 4.6
Opus 4.6 is the most capable model in the Claude 4 family. It shares the 1-million-token context window with Sonnet and can produce longer outputs — up to the equivalent of a very long report or book chapter in a single response.
Its defining feature is the depth of its reasoning. For multi-step, complex tasks — the kind where getting the right answer requires carefully working through many pieces of information in sequence — Opus thinks more carefully before responding. It's not just "more powerful" in a vague sense; it's specifically better at problems that require sustained, careful reasoning rather than quick recall or pattern-matching.
Opus 4.6 is also dramatically cheaper than its predecessor. Opus 4.1 cost $15 per million input tokens; Opus 4.6 costs $5 — a 67% price reduction. That kind of drop signals Anthropic's growing efficiency and their intent to make frontier-level capabilities more accessible.
Who this is for: Power users and developers tackling genuinely hard problems — complex research, advanced coding tasks, multi-step workflows where an AI needs to make a series of decisions. For everyday tasks, it's overkill. For the hard stuff, it's worth it.
How You Access Claude
Claude is available through several different access points, and the right one depends on what you're trying to do.
Claude.ai is the consumer web app and the easiest starting point for anyone. There's a free tier with usage limits, a Pro plan at $20/month that gives you higher limits and access to Opus, and a Team plan for organizations that need shared access and admin controls. Sonnet and Opus also support image and document inputs — you can upload a PDF, a chart, or a photo and Claude can read and respond to it.
The Claude API is for developers who want to build Claude into their own products or workflows. Instead of using the chat interface, they call Claude programmatically and pay per use based on which model they're using and how much text they're sending and receiving.
Claude Code is an AI coding assistant available as a command-line tool and as an extension for code editors like VS Code and JetBrains. It's built for developers who want Claude integrated directly into their coding environment — not a separate chat window, but a tool that can read your code, suggest changes, and help you work through problems without switching context.
Claude Cowork is Anthropic's newest product, currently available as a research preview to a limited set of early-access users. It's a desktop assistant that can create and edit files directly on your computer, with plugins designed for specific professional domains — legal, finance, HR, engineering. It's not something most users can download and try today, but it signals where Anthropic is heading: AI that operates on your local files and integrates into professional workflows, not just a chat window in a browser.
The Future — What's Coming
Legacy models and deprecations
As Claude 4 has become the standard, older models are being wound down. The most immediate change: Claude 3 Haiku is being retired on April 19, 2026. Developers and organizations using that model need to migrate to Haiku 4.5 before that date.
Other Claude 3 and early Claude 4 models remain available but are not recommended for new projects. If you're building something today, start with a Claude 4 model.
Claude Mythos — the leak
In late March 2026, a configuration error in Anthropic's internal content management system accidentally exposed roughly 3,000 unpublished assets — including references to a model called Claude Mythos, also referenced internally as "Capybara."
Anthropic has confirmed the model exists and is in early-access testing. Beyond that, they've been careful about what they've said publicly. What the leaked materials suggested: benchmark scores on coding, academic reasoning, and cybersecurity tasks that are dramatically higher than Opus. A step change, not an incremental improvement.
What's notable isn't just the capability claim — it's the reason Anthropic has given for the slow rollout. They've cited cybersecurity risk concerns specifically, not just the usual "we want to get it right." That framing is significant. It suggests the model is capable enough in sensitive domains that Anthropic is genuinely uncertain about the consequences of releasing it broadly.
No public release date has been announced. What we can say: Anthropic is already working on a tier beyond Opus, it appears to represent a substantial jump in capability, and they're being more cautious than usual about when and how it comes out.
A Moving Target
The Claude model landscape changes fast. What's current in this article reflects the state of Anthropic's lineup as of late March 2026 — check the Last reviewed date at the top of this page to know how fresh this information is, and check Anthropic's official documentation for the most up-to-date specs and pricing.
If you want to keep up with changes like these without doing the research yourself, the AI Decoded weekly covers new model releases and what they actually mean for the way people work.