What's Changed in Claude: A Running Guide

Claude moves fast enough that a feature you learned last month may have moved, been renamed, or been replaced by something better. This page is a running guide to Claude updates: every meaningful change to Chat, Cowork, Code, Skills and the underlying models, dated and written in plain English so you can see what's new without reading a release note built for engineers.

It's for people who use Claude to get real work done and want to keep up without living on the changelog. We review it monthly and add the changes that actually affect how you work, newest first. Small bug fixes and developer-only tweaks are left out on purpose. If a button moved or a whole product launched, it's here.

Last updated: July 2026.

Watch: this running guide is the companion to Jan Meinecke's full Claude Cowork course.

How to use this page

Read from the top for the latest, or skim the dates for the month you last checked in. Each entry says what changed and, more usefully, what it means for you. If you're new to Claude entirely, start with our guide to how to use Claude first, then come back here to stay current.

One thing worth saying up front: the Claude app is now three products in one window, Chat, Cowork and Code, and Anthropic keeps refining how you move between them. If the layout looks different from a tutorial you're following, that's usually a navigation change, not you doing something wrong. The Claude Chat vs Cowork vs Code explainer covers what each one is for.

July 2026

Claude Tag puts Claude in your Slack as a team member. Claude launched Claude Tag, which adds Claude to your Slack workspace as a shared teammate. Anyone in a channel tags @Claude to hand it a task, the whole team sees it work in the thread, and it remembers the channel's context over time. It's the first properly multiplayer Claude: the team shares one Claude instead of each person starting from scratch in their own window, and unlike Cowork or Code it runs asynchronously, including on a schedule, without your laptop open. It's on Team and Enterprise plans, billed on usage, and from 3 August 2026 it replaces the older Claude in Slack app on those plans. If your team lives in Slack, this is the update to act on. Our guide to what Claude Tag is and how to set it up walks through the setup, and Claude for Slack covers how it fits alongside the other ways to connect Claude and Slack.

Claude Opus 5 became the new flagship. On 23 July, Anthropic released Claude Opus 5, its new top model for the hardest reasoning, coding and knowledge work. It comes close to the frontier Fable 5's intelligence at roughly half the price, ships with a 1-million-token context window and a higher "xhigh" reasoning mode, and matches the cost of the Opus 4.8 it replaces while clearly outperforming it. For everyday use the takeaway is simple: when you reach for maximum horsepower in the model picker, Opus 5 is now what you're reaching for, and it's a step up at no extra cost over the old Opus.

Chat and Cowork now live in one place. Claude merged Chat and Cowork into a single view. When you open a new thread you pick Chat or Cowork with a toggle at the top, rather than moving between separate apps. Nothing about the two jobs changed, Chat to think and Cowork to do the work, but the way you switch between them did. If you learned Claude from a walkthrough filmed before this, the products are the same underneath and only the switch moved.

You can now record a skill. Claude added a "Record a skill" option that records your screen, clicks, typing and voice while you do a task once, then turns that recording into a repeatable skill. It works a lot like recording a macro in Excel, except it runs across your apps rather than one spreadsheet. You no longer have to describe a workflow to build a skill from it, you just do the task while Claude watches. One caution Claude shows on screen: don't type passwords or display sensitive information while recording.

Cowork came to web and mobile. Claude Cowork, which had been a desktop-only app, expanded to the web and to iOS and Android in beta, starting with Max subscribers. The new versions run on remote sessions hosted on Anthropic's servers rather than only on your machine. The practical upshot is real: a Cowork task can now keep running, and run on a schedule, with your laptop shut. When Claude hits a decision that needs you, it sends an approval prompt to your phone instead of stopping and waiting. Work you used to have to babysit at your desk can now happen in the background.

Memory changed how it stores things. Claude's memory moved from a single daily summary to a set of individual, categorised entries that it reads and updates as you chat. In practice that means it remembers more precisely and you can open the memory panel, see each thing it has saved grouped by category, and edit or delete any single entry. If you turned memory on months ago, it's worth a look, because it's more useful and more controllable than the version you met.

Claude Fable 5 was pulled, then brought back. Fable 5, the frontier model announced on 9 June, was suspended within days of launch under a US export-control order, then redeployed on 1 July once the order was lifted. Anthropic brought it back with an added safety classifier that blocks the flagged jailbreak in over 99 percent of cases and reroutes those requests to Opus 4.8. For most people this is background noise: Fable 5 is the top of the range and the most expensive model, so you'd only reach for it on the hardest jobs. The practical takeaway is that it's available again, and the everyday Claude you use was never affected.

June 2026

Claude Sonnet 5 launched as the new default. On 30 June, Anthropic released Claude Sonnet 5, built to be its most capable everyday model for agentic work: making plans, using tools like browsers and terminals, and running on its own at a level that recently needed far larger, pricier models. Its performance sits close to the top-tier Opus model at a lower price, and it became the default for Free and Pro users. For most people this is invisible and good: the Claude you open is quietly smarter and faster at multi-step jobs than it was, at no extra cost.

Frontier and flagship models moved up. Alongside Sonnet 5, the model line kept climbing. Claude Opus 4.8 arrived in late May as the heavyweight for the hardest reasoning, and Claude Fable 5 was announced in early June as the new frontier model. You don't need to track version numbers to benefit, but if a task feels like it needs maximum horsepower, the model picker is where you reach for Opus.

Connectors got easier to roll out. Anthropic added zero-touch provisioning for connectors, the feature that lets Claude reach the apps you already use. IT admins can now set a connector up once through the company's identity provider and push it to everyone automatically on first login, and connectors came to mobile in beta. If your organisation has held off because setup was fiddly per person, that objection is largely gone.

April to May 2026

Cowork went generally available. After its January preview, Cowork became available on every paid Claude plan in April, from Pro upward, on macOS and Windows. This is the point it stopped being an experiment and became something you could reasonably put in front of a whole team. If you tried it early and found it rough, the April version is the one to judge it on.

Claude Code got a full desktop redesign. On 14 April, Anthropic redesigned the Claude Code desktop app around running several agents at once: a multi-session sidebar, a drag-and-drop workspace, isolated copies of a project per session so parallel work doesn't collide, and cloud Routines for repeatable jobs. This one is aimed at developers, so most business users can skip the detail. The reason it's here is the pattern: the same Routines and parallel-session ideas tend to reach Cowork next, so it's a preview of where everyday agents are heading. If you do build software, our what is Claude Code guide covers the basics.

Early 2026

Claude Cowork launched. The biggest change of the year for non-technical users. Cowork arrived as a research preview in January, an agent that works on the actual files on your computer and inside your apps, rather than just answering questions in a chat window. It's the product that turned Claude from a clever chatbot into something that does your work. Live Artifacts, Routines and a plugin marketplace followed over the first months of the year, and Chat and Cowork moved toward a single shared interface for projects and artefacts. If you've only ever used the chat box, this is the update to act on. Our Claude Cowork use cases piece shows what people actually do with it.

Memory came to every plan. From March, chat memory became available on the Free, Pro and Max plans rather than paid tiers only. Claude can now carry relevant context between conversations for everyone, so you repeat yourself less.

Late 2025

Skills arrived, then became an open standard. In October 2025, Anthropic launched Claude Skills: a way to teach Claude a whole task once and run it on a single command, shipping first with document-creation skills for slides, spreadsheets, docs and PDFs. In December, the underlying Agent Skills specification was published as an open standard, which means skills aren't locked to Claude alone. Skills are the single most valuable feature for regular users, because they turn a job you do every week into something you delegate. If you build one thing off this whole page, build a skill. Our how to create a Claude skill guide walks through it without code.

Claude Haiku 4.5 shipped. The fast, low-cost model in the line was refreshed in October 2025, which matters most when you're running lots of small jobs or want quick responses over maximum depth.

Why Claude keeps changing under you

It helps to know why this page needs to exist at all. Claude is three products sharing one brain and one window, and Anthropic ships changes almost every week. Menus move, features graduate from beta to default, and models are swapped for better ones behind the same name. None of that is you getting it wrong. The skill that lasts isn't memorising today's menu, it's understanding the shape of the tool: Chat to think, Cowork to do the work, Code to build software, with connectors and skills making all three reach your tools. Learn that shape and every update is a small adjustment rather than a fresh start. It's the difference between chasing features and knowing where you sit on the AI skills ladder.

For the official record, Anthropic keeps its own release notes and a developer-facing platform changelog. Both are thorough and both are written for people who want everything. This page is the opposite: the short list of what actually changes your week.

Common questions about Claude updates

How often is Claude updated? Something ships most weeks, from small fixes to whole products. The changes worth your attention, new features, model upgrades, and interface moves, land every few weeks. We review this page monthly and add those.

Do I have to do anything when Claude updates? Usually not. Model upgrades and most features arrive automatically. The exceptions are the big ones worth acting on yourself: getting on a paid plan, turning on memory, and building your first skill.

Why does my Claude look different from a tutorial? Almost always a navigation or naming change. The products underneath are the same. Match what the tutorial is doing to Chat, Cowork or Code and you'll find the equivalent.

Where's the official changelog? Anthropic publishes full release notes in its help centre and a separate changelog for developers on its platform docs. This page is the plain-English, business-user version of those.

The bottom line

The pattern across every Claude update this year is the same: Claude is moving from a chatbot you ask to an agent you delegate to. Cowork went from a January preview to running scheduled work on your phone with the laptop shut. Skills became an open standard. The default model quietly got smarter twice. You don't need to track version numbers to keep up, you need to know the three products, turn on memory, and build a skill or two. Do that and each new update is a bonus, not a scramble.

We keep this page current so you don't have to read the release notes. Bookmark it and check back monthly.

If you'd rather your team learned this properly and stopped relearning the menu every quarter, our hands-on Claude Cowork training takes people from their first task to a working set of skills that keep paying off no matter what changes.