Claude by Anthropic

Claude runs on three model tiers in 2026. Claude Haiku 4.5 is the fastest and lightest, built for high-volume, quick-turnaround tasks with a 200K token context window. Claude Sonnet 4.6 is the everyday workhorse, the default on most plans, with a 1 million token context window and near-Opus quality at a fraction of the cost. Claude Opus 4.8 is the frontier model, best for complex reasoning, long-horizon planning, and tasks where quality matters more than speed. Thinking modes vary by model: Sonnet 4.6 supports both Extended Thinking and Adaptive Thinking, Opus 4.8 supports Adaptive Thinking only, and Haiku 4.5 supports Extended Thinking only. Claude is consistently rated the best AI for writing, preferred in blind evaluations 47% of the time versus 29% for GPT-5.4.

Extended Thinking lets you toggle deep reasoning on, so Claude thinks before responding with its thought process visible. It is supported on Sonnet 4.6 and Haiku 4.5. Adaptive Thinking is where Claude decides for itself whether to think deeply based on the complexity of the request, no toggle needed. It is supported on Sonnet 4.6 and Opus 4.8. So Sonnet 4.6 supports both modes, Opus 4.8 supports Adaptive Thinking only, and Haiku 4.5 supports Extended Thinking only. Both modes produce noticeably better results on multi-step problems, complex analysis, and anything where getting the framing right matters as much as the answer.

Claude Sonnet 4.6 and Opus 4.8 support a 1 million token context window, roughly 750,000 words, or the entire Harry Potter series plus Lord of the Rings in a single conversation. Haiku 4.5 supports 200K tokens. In practice, you can upload an entire business strategy document, a full year of customer feedback, or a complete legal contract and ask Claude to reason across all of it at once. It will not forget earlier parts or lose context mid-conversation. This is one of Claude's most distinctive technical advantages over competitors in 2026.

Projects are Claude's most powerful organisational feature. Each Project is a persistent workspace holding uploaded files, custom instructions, and conversation history, all carried across every session inside it. Open a Project conversation and Claude already knows the context: what you are building, what your standards are, and what files it has access to. Projects also have their own separate memory, so context inside a Project stays focused and does not bleed into other work. On Team plans, Projects can be shared so the whole team works from the same context.

Memory lets Claude summarise your conversations and carry key context across sessions: your name, role, preferences, working style, and ongoing projects. It updates automatically every 24 hours and is fully transparent. You can view, edit, or delete anything Claude has remembered at any time. Memory and Projects work as a pair. Memory handles who you are and your broad preferences. Projects handle the specific context for a body of work. As of March 2026, memory from chat history is available to all users, including Free, on web, desktop, and mobile.

Claude Team plan unlocks shared Projects, centralised billing, admin controls, and usage visibility. This shifts Claude from a personal productivity tool to a shared team capability. The whole team works from the same uploaded files, the same custom instructions, and the same AI configuration. Team data is never used for model training by default. On Business and Enterprise plans, admins can manage access, set permissions, enable or restrict specific features, and connect organisation-wide data sources.

Web Search is a server-side tool built directly into Claude's reasoning loop, not a browser plugin. Toggle it on in claude.ai and Claude decides for itself whether a query needs fresh data, runs a targeted search, and integrates the results into its response with citations. Unlike a standard search engine, Claude does not just return links. It reasons over the results and synthesises them into a direct answer. For time-sensitive queries, prompt "search the web for current information on X" so Claude pulls live data rather than relying on training knowledge.

Research is Claude's deep research mode. It operates agentically, running multiple web searches that build on each other, exploring different angles automatically, and producing a comprehensive cited report in minutes. It needs Web Search turned on. Unlike a single search, Research runs an iterative loop: it plans what to investigate, searches, reads, finds gaps, searches again, and synthesises across all sources. It can also search your connected Google Workspace alongside the public web. The result is a structured, verifiable report rather than a quick answer.

Claude accepts PDFs, Word documents, spreadsheets, images, and other file types directly in conversation. With a 1 million token context window, it can process entire document libraries, lengthy contracts, or large datasets in a single session. Attach files to a Project to make them permanently available without re-uploading. Claude can summarise, compare, extract, rewrite, audit, and reason across everything you upload, with significantly better document comprehension than most competing models thanks to its high-resolution vision support (up to 2,576 pixels per image).

Artifacts are Claude's output layer, a live side panel that renders interactive content instead of returning plain text. Ask Claude to build something visual, interactive, or structured and it appears in Artifacts as a working preview you can interact with, edit, share via link, or embed on a website. Artifacts support React components, HTML/CSS/JS apps, SVG diagrams, Mermaid flowcharts, code files, and formatted documents. As of April 2026, Live Artifacts in Cowork connect to MCP servers and refresh with real-time data every time you open them, turning a static snapshot into a dynamic, always-current tool. Anyone can view a shared Artifact via link, even without a Claude account.

Claude Design is a dedicated visual creation workspace powered by Opus 4.8. Launched April 2026 in research preview. Describe what you want, a mobile app prototype, a pitch deck, a one-pager, a marketing visual, and Claude generates a first version on a live canvas. Refine it via conversation, inline comments, direct edits, and custom adjustment sliders. Export as PDF, PPTX, URL, or send directly to Canva for final polish. It can read your company's design system and apply it consistently across every project it creates. It is not a full replacement for Figma or Canva. It is the fastest path from an idea to something visual, particularly for non-designers who need to communicate an idea before a designer touches it.

Skills are reusable instruction packages that teach Claude exactly how to execute a specific task: your standards, your format, your process. A Skill is a folder containing a SKILL.md file with instructions, plus optional supporting files, templates, and scripts. Claude loads only the Skills relevant to the current task, so performance stays sharp no matter how many are installed. Skills are composable, so Claude coordinates multiple Skills automatically when a task needs them. They work across claude.ai, Claude Desktop, Claude Code, and Cowork, and are available on all plans as of April 2026. Skills can also be shared, and Anthropic ships pre-built Skills for Excel, PowerPoint, Word, and PDF workflows.

MCP (Model Context Protocol) is an open standard developed by Anthropic that creates secure bridges between Claude and external tools and services. The Connectors Directory now has over 375 integrations, including Google Drive, Gmail, Slack, Notion, HubSpot, Jira, Salesforce, Stripe, GitHub, Adobe Creative Cloud, Canva, Asana, and many more. Claude inherits your existing permissions in each connected tool, so it can only see what you can already see. Some connectors are read-only. Others are interactive, letting Claude take actions like sending a Slack message or creating a Jira ticket. Free users get one custom connector. Paid plans unlock full custom MCP server connections.

Claude Apps are a subset of MCP Connectors that go beyond data retrieval. They render interactive UI directly inside the chat. Where a standard connector lets Claude read your Google Drive, a Claude App lets you open Asana and update a task timeline, draft and preview a Slack message, or interact with a Canva design without leaving the Claude interface. Available apps include Figma, Canva, Asana, Slack, Adobe, Box, monday.com, and others. Find the directory at claude.ai/directory.

Cowork is Claude's desktop agent, the third tab in Claude Desktop, available on macOS and Windows. Launched January 2026, generally available April 2026. Unlike the chat interface, Cowork has explicit permission to read, edit, and create files in folders you specify on your local computer, run multi-step tasks autonomously, run scheduled recurring tasks, and control your computer screen (clicking, typing, navigating) when no connector exists for the tool you need. It requires the Claude Desktop app and a paid plan. It is the non-developer version of Claude Code, same agentic architecture, no terminal required. Cowork has its own persistent memory across sessions.

Computer Use lets Claude interact directly with your desktop, applications, and browser, clicking buttons, filling forms, navigating interfaces, as a fallback when no MCP connector exists for the tool in question. It operates inside Cowork and Claude Code, with no sandboxing between Claude and your applications. It is currently in research preview for Pro and Max plans on macOS and Windows. Because Claude interacts directly with your real desktop, not a virtual machine, it is the most powerful and the most permission-sensitive feature in the product. Use it for tasks where you are comfortable with Claude acting on your behalf.

Claude Code is Anthropic's terminal-based autonomous coding agent, among the strongest generally available models on SWE-bench Verified (88.6%) in 2026. It reads your entire codebase (up to 1 million tokens of context), plans a sequence of changes across multiple files, writes the code, runs tests, handles failures, iterates, and commits the result. Developers define the goal and review the output. The execution loop runs independently. Available via terminal CLI, VS Code and JetBrains IDE extensions, Claude Desktop, and Claude Mobile. It also includes Agent Teams, multiple Claude Code instances working on different parts of a problem in parallel, coordinated by a lead agent. For non-technical participants, this is the equivalent of having a developer on call 24/7 who works while you sleep.

Claude is now available as native add-ins inside Microsoft 365 applications. Claude in Excel (generally available October 2025) provides workbook-wide understanding, formula assistance, data analysis, and assumption updates with cell-level citations. Claude for Word (generally available April 2026) handles drafting, editing, and summarising directly in the document. Claude in PowerPoint assists with deck creation, slide rewrites, and speaker notes. Claude in Outlook (public beta May 2026) handles email drafting, summarising, and response suggestions. Available on Pro, Max, Team, and Enterprise plans.

Claude is available as a native desktop app (macOS and Windows) and mobile app (iOS and Android). The desktop app is where Cowork and Computer Use live. It is the most powerful surface for multi-step, file-based, and agentic work. The mobile app supports full Claude functionality including Projects, Memory, MCP Connectors, and voice input. All surfaces share the same conversation history and project context. Claude Code is also accessible from the mobile app, letting you check in on a running coding session from your phone.