Microsoft 365 edition
Microsoft explains what your Copilot licence covers and what will cost extra per use, two new models are joining the model picker, and Excel finally shows you exactly what it changed.
Copilot gets two bills: your fixed licence and credits for agent work
Chat, Word, Excel and PowerPoint stay in your licence, just as before. Long agent tasks like Cowork and Autopilot will soon be billed per use.
On 25 September 2026, Microsoft published details on how Copilot will be billed from now on. There will be two separate pots. The first pot is the fixed user licence, the Copilot User Subscription License (USL). This covers your daily work: Copilot Chat and Copilot in Word, Excel and PowerPoint. Your employer buys this licence per employee. Whether you use Copilot a lot or a little, the bill stays the same. It includes the models GPT-5.6 and Claude Sonnet 5. Newer models like Opus 5 will be added, but with limits. The second pot is new and is called Copilot Credits. This is usage-based: you pay for what you use. It covers the heavier tasks that an agent carries out on its own for hours, such as Cowork, Code and Autopilot. Important: these usage-based services are switched off by default. An administrator first needs to set a spending policy in the Microsoft 365 admin centre, with budgets per organisation, per group and per user. So if you don't see Cowork or Autopilot, there's probably just no budget set yet. This week, ask your administrator whether a spending policy is already in place and what budget your team has. The difference lies in how long the model keeps working. If you ask Copilot in Word to rewrite a paragraph, that's one round: read, write, done. If you set Autopilot or Cowork on a task, the model performs dozens of steps in a row: looking things up, combining, checking, trying again. Each step costs computing power, and that varies a lot between colleagues. Credits are a counter on that computing work. The spending policy is the tap in front of it: without a budget, nothing happens, and with a budget per group and user, you know in advance what the maximum can be. As long as everything was in the fixed licence, you could experiment freely: a failed attempt cost nothing extra. With credits, a long agent task that you then throw away does cost money. This can go wrong in two ways: people become afraid to try anything, or someone sets an agent on a task that was really five minutes of manual work. So agree in advance what work an agent is allowed to do, and who keeps an eye on the counter.
Background for subscribersThis week you'll see Copilot shift from one fixed amount to a system with two pots: your fixed licence for daily work, and credits you pay per use for heavy agent work like Autopilot and Cowork. New, heavier models are also being added to the model picker, and Copilot in Excel now shows with one click exactly which cell it changed. For you, this mainly means: check whether a budget has already been set before you start using agent features, and always check for yourself whether a Copilot answer is actually correct.
Claude Opus 5.5 and GPT-6 Sol are coming to Copilot's model picker
On its Copilot community page, Microsoft says that Claude Opus 5.5 and GPT-6 Sol are being added to the model picker in Microsoft Copilot. You'll be able to choose which model works on a task, within Word, Excel, PowerPoint and Cowork. This fits a broader trend: organisations are using agents more and more in real work processes. In the same post, Microsoft also announces an 'AI at Work' webinar in October, about Copilot, apps and agents. Good to know: the model picker is part of your employer's paid Microsoft 365 Copilot environment. Whether you see the new names depends on what your administrator has enabled. Microsoft is rolling this out in phases, so your screen may lag behind this announcement. This week, check the prompt box to see if new model names have appeared, and try one out on a task you can judge well. A model picker works like choosing between colleagues with different talents: one model is faster and cheaper, another thinks through more steps and costs more computing power. Your question goes to the model you selected, the rest of Copilot stays the same: same files, same permissions, same buttons. Only who writes the answer changes. That's why you'll notice the difference mostly with long, complex pieces, and hardly at all with a short summary. More choice also means more chance of the wrong choice: picking a heavy model for a standard email wastes time and uses up your Copilot budget faster. Each model also writes slightly differently, so if you switch partway through a long document, the tone becomes uneven. It's better to agree on two fixed choices within your team, one for quick work and one for pieces that go outside the organisation.
Background for subscribersThis is a taster.
Subscribers read the whole Microsoft 365 edition: every story, each with its background in plain language.
Subscribe →Not ready for that? Read the free letter first →
AI and tech, made simple.
This is the Microsoft 365 edition. Subscribers get the background to every story, in plain language.
Subscribe
inklaretaal · Amsterdam · © 2026