Tool updates Claude

Tool notes · Claude · Issue 1 · November 2026

Claude this spring: cheaper models, Cowork in every chat, and credits to claim

Anthropic has shipped a new family of models, folded Cowork into ordinary chats and started testing an agent that works while you are away. Here is what matters for a business, covering September and October.

By Karrina Mountfort5 November 20264 min read7 entries

Claude has had a busy spring. A new family of models arrived over three weeks, Cowork became part of every conversation, and Anthropic opened a preview of Duet, an agent that works across your apps on its own.

The thread running through all of it: Claude is doing more by itself. That makes the setup around it, which connectors it has, what it is allowed to change and who checks its work, more important each month.

Entries are numbered so the checklist can point to them.

If you only have 2 minutes

  1. 1New 5.5 models. Opus 5.5 costs 40% less to run than Opus 5, and Sonnet and Haiku 5.5 are faster and cheaper again.
  2. 2Credits to claim. Team and Max plans now include monthly API credits, but someone has to claim them.
  3. 3Cowork in every chat. On Pro and Max, a quick question can become a bigger task without switching screens.
  4. 4Duet is in preview. A proactive agent across your connected apps. Interesting, not for real work yet.
  5. 5Enterprise gets smart reports. See cost, sticking points and repeated work worth turning into skills.

Checklist

Do first

  • On a Team or Max plan, claim your monthly API credits · Now · see 2
  • Review which connectors Claude has, and what they can change · Before using Cowork on real work · see 3

When you have time

  • Match the model to the job · see 1
  • On Enterprise, read the smart report with your team leads · see 6
  • Check the sensitive topics setting for memory · see 5

Models and cost

1.The 5.5 models: Opus, Sonnet and Haiku

Who: Everyone using Claude · Rollout: 22 September to 7 October · Where: Claude apps and API

Opus 5.5 arrived on 22 September, and Anthropic says it costs 40% less to run than Opus 5. Sonnet 5.5 followed on 28 September as the faster, lower-cost option, and Haiku 5.5 on 7 October, described as the cheapest, fastest and most capable small model Anthropic has released.

Why this matters

The smaller models are now good enough for a lot of everyday work. If you pay for usage, choosing well adds up.

  • In plain words: Three new models: one for hard thinking, one for everyday work, one for quick jobs.
  • A real-world example: An insurance broker uses Opus for reading long policy wordings, Sonnet for drafting client emails, and Haiku to sort and tag incoming enquiries.
  • Our take: Do not default to the biggest model for everything. Pick by task, and check the cheaper one first.
Which model for which job
Opus 5.5Hard thinking · Reading long policy wordings
Sonnet 5.5Everyday work · Drafting client emails
Haiku 5.5Quick jobs · Sorting incoming enquiries
Illustration by Webbased AI of matching models to tasks. Not an Anthropic screenshot.

Claude Help Center: Release notes

2.Monthly API credits on Team and Max plans

Who: Team and Max plan admins · Rollout: 7 October 2026 · Where: Billing settings

Max and Team plans now include monthly API credits. They are claimed by linking a Claude Console organisation in billing settings.

Why this matters

Credits you are already paying for, useful for testing connectors and small automations. They are not applied automatically.

  • In plain words: Free API budget each month, if you claim it.
  • A real-world example: A small software reseller uses its credits to test a connector to its job system before paying for anything more.
  • Our take: Claim them. It takes minutes, and it is the cheapest way to try an automation properly.

Claude Help Center: Release notes

Claude doing more

3.Cowork in every chat

Who: Pro and Max plans · Rollout: Rolling out from 16 September · Where: Claude on web, desktop and mobile

Cowork is now available from any conversation. Chats, tasks, projects, connectors and skills carry across, so you can ask a quick question or hand Claude a bigger task in the same place.

Why this matters

When Claude can act, not just answer, its connector permissions matter as much as its answers.

  • In plain words: Claude can take on a whole job from the same chat you were already in.
  • A real-world example: An architect asks Claude a question about a council requirement, then hands it the task of checking a set of plans against it, in one conversation.
  • Our take: Before using it on real work, check which connectors Claude has and what they can change. Start read-only, and switch on more once it has earned it.
Copilot · Home

Home

Agents

Skills

Connectors

Projects

What can I help with?

Ask

Hand off a task

Ask a question or describe a task…
Illustration by Webbased AI of asking and handing off in one place. Not an Anthropic screenshot.

Claude Help Center: Release notes

4.Duet, a proactive agent, in research preview

Who: Pro and Max plans · Rollout: 21 October 2026, research preview · Where: Duet app

Anthropic opened a research preview of Duet, a new app where Claude works as a proactive agent across your connected apps.

Why this matters

It is a look at where assistants are heading: agents that keep working while you are away.

  • In plain words: A Claude that gets on with things in the background.
  • A real-world example: Fine to explore on a personal account. Not one to point at customer data yet.
  • Our take: Nothing to do. We are trying it, and will write it up when it is ready for business use.

Claude Help Center: Release notes

5.Memory across chat and Cowork

Who: Paid plans · Rollout: 25 August 2026 · Where: Claude settings

Memory now works across chat and cloud Cowork, with editable topics. Sensitive topics stay out of memory unless “Include sensitive topics in memory” is switched on.

Why this matters

Memory saves re-explaining, but it should remember work, not personal details.

  • In plain words: Claude remembers context between chats, and you can see and edit what it remembers.
  • A real-world example: A team member checks the memory topics and removes a client detail they did not want kept.
  • Our take: Leave the sensitive topics setting off unless there is a clear reason, and show people where to see what Claude remembers.

Claude Help Center: Release notes

For admins

6.Smart reports for Enterprise

Who: Enterprise admins · Rollout: 10 September 2026, beta · Where: Admin

Smart reports show how a team uses Claude, including cost, where people get stuck, and repeated patterns worth packaging as shared skills.

Why this matters

It turns “are people using it?” into a list of what to train and what to build.

  • In plain words: A report on how your team uses Claude, and where to improve.
  • A real-world example: A report shows five people each writing their own version of the same weekly summary. That becomes one shared skill.
  • Our take: Read it with your team leads and use it to aim upskilling, not to rank people.

Claude Help Center: Release notes

7.Catching up: Microsoft 365 write access and connector permissions

Who: Admins · Rollout: May to July 2026 · Where: Connectors and admin roles

Since 7 July, the Microsoft 365 connector can draft and send email, manage calendar events and edit OneDrive and SharePoint files when write tools are enabled. Teams stays read-only. A Microsoft Entra administrator must consent to the new permissions.

Since May, admins can control which connectors and individual tools each custom role can use.

Why this matters

This is the difference between Claude reading your email and Claude sending it.

  • In plain words: Claude can now act in Microsoft 365, if you let it, and you can control who gets which tools.
  • A real-world example: A practice lets its office manager use calendar write tools, and keeps everyone else read-only.
  • Our take: Turn write tools on deliberately, for the roles that need them, with a person checking anything that goes to a customer.

Claude Help Center: Release notes

Watching next month

  • Duet as it moves beyond research preview
  • Cowork reaching more plans
  • Any change to pricing or plan limits

How this issue was put together

We read Anthropic’s release notes and only include changes with a date, a cost or something you need to do.

Every fact links to its source. The examples are made-up New Zealand businesses, not clients. The figures are our own illustrations, not Anthropic screenshots.

Official resources

Want this handled, not just read?

We set up, connect and look after Copilot and the other AI tools, so changes like these are taken care of.

How we roll out AI tools