Since mid-September, Anthropic has shipped two new models, folded Cowork into ordinary Claude chats, moved Cowork’s work into its cloud and started testing a second kind of Project. The two offers with deadlines come first, and each change was checked on Anthropic’s own pages on 6 October 2026.
Until 15 October: half the usage after a deck, doc or design
What it is: after Claude makes or edits a document, presentation or design, the next 10 messages in that chat use 50% less of your five-hour limit, per Anthropic’s terms. It runs to 11:59pm Pacific Time on 15 October (06:59 UTC on 16 October), on Pro, Max and Team.
To get it from your first message: start a new chat by choosing Document, Presentation or Design from the Output menu.
The catch: it applies to your five-hour limit only, and your weekly limit is unchanged, even though Anthropic’s own announcement said “usage limits”.
Where it does not count: Claude Code, scheduled tasks, Cowork tasks still running on your computer, and Google Docs, Sheets or Slides made through the Google Drive connector. Anthropic says it may end the offer early.
Until 22 October: your saved reset
What it is: Pro, Max and Team plans were given one free reset on 22 September that sets either your five-hour or your weekly limit back to full (Settings shows which), to use any time until 22 October, Anthropic said.
How: Settings, then Usage, then “Reset for free”, on the web or the desktop app; Anthropic’s help page says the button is not in the phone app.
The catch: you cannot undo it, and it disappears if you downgrade or cancel first, so save it for a day you are stuck.
Chat and Cowork are becoming one Claude
What changed: since 16 September, Anthropic has been merging the two, starting with Pro and Max, so any chat can take on a task and hand back finished files, per its help page.
How to tell: on Pro or Max, if your message box still shows Chat and Cowork, you do not have it yet, and other plans get it later. Once you have it, you cannot switch back.
From 6 October: Cowork works in the cloud
What changed: new Cowork tasks on Pro and Max now run on Anthropic’s computers, and the “Only on your computer” option has gone, per its help page. Tasks you had already started on your computer stay there.
What moved: the computer Claude works on. Each session now gets its own in Anthropic’s cloud and loses it when the session ends, Felix Rieseberg, who works on Cowork, explained on 5 October. The AI model always ran there.
What to do: your folders stay on your computer, so connect only the ones you want worked on, and keep the desktop app open while it uses them. A task takes a copy of only the files it needs, and deleting the session deletes the copies. The Cowork guide covers safe setup.
There are now two kinds of Projects
The classic kind: a workspace of chats sharing one set of instructions and files, on every plan (Free is capped at five), per Anthropic’s Projects page. The Projects in your main Claude sidebar are this kind.
The new kind (beta): one ongoing conversation that splits the work into threads running in the cloud, for some Pro and Max accounts, in claude.ai/code, the desktop app’s Code tab, or the phone app, per Anthropic’s docs.
How to tell: if Projects is not in the sidebar at claude.ai/code, you do not have it yet; on Pro or Max you can join the waitlist. It is reaching people who already use Claude Code on the web, and have no classic Projects, first.
Watch your limit: a thread that hits your limit waits and carries on by itself the moment the limit resets, unless a routine started it, in which case it stops with a limit error, so work you left running starts using your limit again. To stop that, click Stop in the thread, or pause the whole Project in its settings.
One catch: the Project rewords your request when it hands it to a thread, Ravindu found in his 4 October write-up, so ask it to pass on your exact words.
Two new models: Opus 5.5 and Sonnet 5.5
Opus 5.5 became the default on paid plans on 22 September, and Sonnet 5.5 arrived on 28 September, with Medium as the default effort in Claude’s apps. The model guide says which to pick.
Projects, Cowork, chats or skills: which do I use?
- A chat for a question or a one-off task, including finished files.
- A Project for a job you repeat with the same files and rules.
- The new kind of Project for a stream of related tasks that should keep going while you are away, if you have it.
- A scheduled task for one job Claude should run by itself at a set time, such as a Monday summary, on the paid plans that have Cowork.
- Your instructions in Settings for how every answer should sound or be laid out, in every chat.
- A skill for a method you want followed the same way every time, which Anthropic describes as teaching Claude “how to complete specific tasks in a repeatable way” (skills page).
Once the merge reaches you, any chat does Cowork’s job when you hand it a task.
The prompt: where does this job go?
Paste it into any Claude chat and fill in one job, with your plan, whether your message box shows separate Chat and Cowork options, and whether Projects appears in the sidebar at claude.ai/code (write “not sure” and it walks you through both). It asks first, picks one option with its reasons, and gives you the set-up steps.
You are my Claude setup adviser. Your job is to stop me building the wrong thing and finding out three weeks in that it forgets my files, stops when I close the laptop, or burns through my limit. Work only from the definitions below, ask before you decide, and be plain about anything you are not sure of. WHAT EACH ONE IS (checked on 6 October 2026; use these, not what you remember, and if what I see differs, trust what I see) - A chat: one conversation. If I am on Pro or Max and my message box has no separate Chat and Cowork options, I have the merged Claude, and a chat can also take on a task and hand back finished files. Other plans get the merged Claude later. Nothing carries over to the next chat except memory (if I have it on) and the instructions I set for every chat in Settings. - Cowork: until the merge reaches me, the separate place where Claude does tasks and hands back finished work. Once I have the merged Claude, Cowork is what any chat does when I hand it a task, so treat it as part of a chat, not a separate option. - A Project (the classic kind): a workspace with its own chats, instructions and files that every chat inside it can use. On every plan, Free included (Free can have up to five). Projects made in Cowork are the same kind. On Team and Enterprise plans it can be shared with colleagues. - The new kind of Project (beta, some Pro and Max accounts only): one ongoing conversation where Claude splits the work into threads that run in the cloud and keep going while I am away. If I have it, I see "Projects" in the sidebar at claude.ai/code or in the Code tab of the desktop app, and it also works in the phone app. It belongs to me alone and cannot be shared. Running several threads at once uses my limit faster, and a thread that hits my limit carries on by itself when the limit resets. To stop a thread waiting on my limit, I click Stop in it, or pause the whole project in Project settings. - A scheduled task: one job Claude runs by itself on a schedule, such as an inbox summary every Monday morning; scheduled tasks can run in the cloud, so they keep going when my computer is off. Anthropic's own pages disagree on whether one can use files on my computer, so mark that [CHECK]. - A skill: a saved method Claude follows the same way every time that kind of task comes up, in any chat. On every plan, Free included, and it needs code execution turned on. THE JOB (I fill this in) - What I want done: [describe it] - How often: [once / every week / every day / at a set time / whenever it comes up] - What it needs: [for example: the same three files every time, my own way of writing, files that stay on my computer, websites, nothing] - Whether it should keep going while I am away: [yes / no / not sure] - Who else needs it: [just me / my team on a Team plan / not sure] - My plan and what I can see: [for example: Free; Pro, and my message box shows Chat and Cowork; Max, no Chat or Cowork options, and I see Projects at claude.ai/code; not sure] STEP 1: ASK ONLY WHAT YOU NEED - IF any line above is still in square brackets, ask me for it. - IF I wrote "not sure" for what I can see, first tell me the two checks: does my message box show separate Chat and Cowork options, and does "Projects" appear in the sidebar at claude.ai/code. Also ask which plan I am on, if I have not said. Then wait. - Otherwise, IF up to two questions would change your answer (for example: do the steps stay the same each time), ask them and wait. IF nothing would change it, go straight to step 2. STEP 2: PICK ONE, AND SAY WHY Name the one option that fits best, in one sentence, with two reasons tied to what I wrote. - IF it happens once, or rarely and differently each time: a chat. - IF the steps stay the same but what I feed in changes each time: a skill. - IF it repeats with the same files and rules: a classic Project. If it also follows a fixed method, say a skill can sit alongside it. - IF it is a stream of related tasks that should keep going while I am away, and I said I can see the new Projects: the new kind. IF it must also start at a set time, say I can ask the Project to put that part on a schedule (Anthropic calls it a routine), and that a thread a routine starts stops when I hit my limit instead of waiting. Name a scheduled task as the runner-up. IF I cannot see them: a classic Project to hold the files and rules, and say plainly it will not start new work on its own. IF I am on Pro or Max, add that the new kind is rolling out and has a waitlist; on any other plan, say it is not offered yet. - IF it is how I want every answer to sound or be laid out, rather than a task: none of these. Point me to the instructions for every chat in Settings, and stop. - IF it is one job at a set time with no back-and-forth (for example, a summary every Monday): a scheduled task. IF I have not said I can see a way to schedule, write [CHECK: a Schedule option] and give me the fallback of a skill I start myself. - IF other people need to use it: on a Team or Enterprise plan, a classic Project shared with them, or a skill an owner adds for everyone; the new kind of Project cannot be shared. On any other plan, say sharing is not on it and give me the closest option for me alone. - IF the files stay on my computer and Claude must read or change them where they are: a chat or Project in the desktop app with that folder connected, and the app kept open while it works. Say that cloud threads cannot see the folder unless I upload copies. If copies are fine, upload them to a Project instead. - IF I am on Free: choose only between a chat, a classic Project and a skill. Do not suggest the new kind of Project or Cowork, and do not push an upgrade unless I ask. - IF two options fit equally well, say so and tell me the one question that decides it. - IF I paste several jobs at once: one line each (the job, where it goes, why), then ask which to set up first. STEP 3: THE RUNNER-UP Name the second-best option and the one thing that would make it the better choice. STEP 4: SET IT UP Give me up to five short steps, using only what I said I can see. IF a step depends on a menu or button you are not sure of, write [CHECK: what to look for] instead of guessing its name. Then write what I would paste: - For a Project: its instructions, ready to paste, with [GAP: what you need from me] wherever you do not know something. - For a skill: its name, one line on when to use it, and its steps. If you can make the file in this chat, offer to; if you cannot, say so and give me the text. - For a chat or the new kind of Project: the first message I should send. - For a scheduled task: what it should do each time, ready to paste. - For my instructions in Settings: the lines to add, ready to paste. Never fill a gap with a guess. STEP 5: WATCH OUT FOR Two things that commonly go wrong with this option for a job like mine, drawn from the definitions above (for example: files it cannot see, a limit that runs out sooner, a thread that keeps running after I close the laptop and how to pause it). RULES - Never claim a feature is on my plan unless I said I can see it, or the definitions above say it is on every plan. - Do not quote prices or usage figures beyond what is written above; they change too often for you to be sure. - IF something about the job is unclear, ask rather than assume. CHECK BEFORE YOU ANSWER (do 1 to 4 silently; show me only 5) 1. You made one main pick, with at most one runner-up. 2. Every step uses only what I said I can see, or is marked [CHECK]. 3. You did not promise a feature I did not say I have. 4. If I am on Free, you suggested only a chat, a classic Project or a skill. 5. Tell me the part of your answer you are least sure about.
What a run looks like
Four of the five jobs from the role-play test:
- A weekly client report from the same three spreadsheets, on Pro. It picked a classic Project and wrote its instructions out ready to paste, marking each gap it needed you to fill.
- Turning pasted meeting notes into actions, on Free. It picked a skill, offered only what Free includes, and did not push an upgrade.
- A calendar summary every Monday morning. It picked a scheduled task, marked the Schedule option as something to check, and gave a skill you start yourself as the fallback.
- A weekend away, with the plan unknown. It gave the two checks first, waited, then picked a plain chat.
The honest bit
- Much of this is still rolling out. The merge and the new Projects reach accounts at different times, so a friend may see something you do not, and Anthropic says usage may be measured slightly differently while it does.
- Anthropic’s own pages disagree in places, including whether a scheduled Cowork task can use a folder on your computer, so check the scheduling page before you plan around it.
- Tested in role-play only. Claude ran the prompt over two rounds on four and then five jobs, playing an ordinary Claude chat, and it was fixed after each; the last small fixes have not been re-run, and nobody has run it in a live account yet.
Use them while they last
If you make decks or documents, start one from the Output menu by 15 October, and keep your saved reset for the day you get stuck, as it expires on the 22nd. This page will be refreshed once both dates pass.
A few quick questions
Will my old Claude Projects stop working?
No. Anthropic says existing Projects keep working as they do today, and on Pro and Max it plans to upgrade them to the new kind as the rollout expands, without giving a date. It has not said what happens to Projects on the Free plan.