Claude Opus 5.5 and GPT-6 Sol in Microsoft Copilot: what UK firms should check first

On 22 September Microsoft posted a short announcement called “More Models, One Copilot”. Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6 Sol were rolling out that day across Word, Excel, PowerPoint, Chat, Cowork and Copilot Studio, with availability varying by licence, access and region. Eight days later Microsoft added two more: OpenAI’s GPT-6.1 Sol and Anthropic’s Claude Sonnet 5.5.

That is four new models in eight days, in the same fortnight Microsoft changed how Copilot is priced. Our living list of the models behind Copilot shows where each one sits today. This post is about what a UK firm should check before anyone starts using them.

What arrived

Claude Opus 5.5 is the first model in Anthropic’s Claude 5.5 family. Anthropic says it performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5. That matters because Fable 5.1 only arrived in Copilot at the start of September, as a frontier model in Cowork and Copilot Studio. If Anthropic’s claim holds, Opus 5.5 brings something close to that capability into Word, Excel, PowerPoint and Chat.

GPT-6 Sol is one of two cheaper models OpenAI added to its GPT-6 family in September, after GPT-6 Astra, its top model. OpenAI says it trained Sol with similar methods to Astra and halved its API prices against GPT-5.6 Sol’s promotional pricing.

GPT-6.1 Sol and Claude Sonnet 5.5 followed on 30 September. Microsoft said they would roll out in Cowork and Copilot Studio straight away, and in Word, Excel, PowerPoint and Chat over the following week.

On benchmarks, both vendors publish their own numbers. On AutomationBench, a test of business workflows across apps, Anthropic reports Opus 5.5 at 40.0% against 26.9% for Opus 5. OpenAI reports GPT-6 Sol at 33.2% on its own run of the same benchmark. The two runs are not set up the same way, so treat neither as a league table. They are a sign that the gap between the everyday models and the frontier ones is closing quickly.

One thing that does not carry across is token pricing. Anthropic and OpenAI both cut prices, but inside Microsoft you are not billed per token. You pay through the Copilot licence or in Copilot Credits.

Where to find them, and why the documentation lags

In Chat and the Office apps the new models appear in the model picker once they reach your tenant. In Cowork they appear in the model dropdown. In Copilot Studio a builder chooses them per agent.

Microsoft’s documentation has not kept up. When we checked on 3 October, the Learn page listing Cowork’s models still showed GPT-6 Astra, Fable 5.1 and Opus 5 but none of the four new models, and the Copilot Studio model table still listed Claude Opus 4.7 as its newest Anthropic model. Go by the list in front of you, which reflects what your administrator has allowed, rather than the documentation.

Check one: is Anthropic switched on, and should it be?

Opus 5.5 and Sonnet 5.5 are Anthropic models, so in a UK tenant they appear only once an administrator has enabled Anthropic. It ships switched off in the UK, the EU and EFTA because Anthropic models are excluded from the EU Data Boundary and, where they apply, from in-country processing commitments.

Anthropic works as a Microsoft subprocessor, so Microsoft’s Product Terms and Data Protection Addendum apply. What you give up is the geographic guarantee. That is a decision for whoever owns client confidentiality, not for whoever happens to be signed in to the admin centre. Our guide to using Copilot with client data sets out the questions, and the steps to enable it are on the living list.

If you decided this when Claude Opus 5 arrived in July, nothing about the decision has changed. The model behind the switch is stronger, and there are more of them.

Check two: the OpenAI setting most firms have never looked at

Since July, Microsoft has offered some GPT models run by OpenAI itself as a Microsoft subprocessor, alongside the GPT models it runs in Azure. Since 24 July those OpenAI-run models have been on for all users at eligible commercial customers unless an administrator turns them off.

Microsoft’s Product Terms and Data Protection Addendum apply to them, and they sit inside the EU Data Boundary apart from exceptions Microsoft documents. But they are currently excluded from in-country processing commitments. In Cowork, Microsoft marks GPT-5.6 Sol, GPT-5.6 Terra and GPT-6 Astra as provided by OpenAI. It had not yet documented how GPT-6 Sol and GPT-6.1 Sol are delivered when we checked.

If your firm has told clients where their data is processed, read the setting under Copilot, Settings, View all, then AI providers operating as Microsoft subprocessors. It is easy to miss, because nobody had to switch it on.

Check three: which side of the billing line

Microsoft changed how Copilot is paid for on 25 September. Everyday AI sits in the per user licence and advanced AI is paid in Copilot Credits.

• Cowork: every task is billed in credits, whichever model runs it. A stronger model makes bigger jobs tempting, so set a spending limit before anyone tries one.
• Copilot Studio: agents bill in credits. On the standard harness Microsoft’s rates are the same for every model it provides, with a premium rate on top for reasoning models, and use by licensed members of your team is included within fair use. On the GitHub Copilot harness everything bills in credits.
• Chat and the Office apps: the licence. Microsoft says GPT-5.6 and Sonnet 5 are included under fair use, Opus models will be included with usage limits and GPT-6.1 Sol and Sonnet 5.5 arrive as part of the licence with limits. Its 22 September announcement did not say how Opus 5.5 and GPT-6 Sol are counted.

We set out the whole split, and the 1 December change for Copilot Business, in Copilot model choice: what the licence covers and what costs Copilot Credits.

Check four: agents you have already built

In Copilot Studio the model is chosen per agent, so nothing you have built changes because a new model arrived. The exception is an agent left on the default model, which Microsoft upgrades from time to time.

Before moving a working agent to Opus 5.5 or GPT-6 Sol, run it against a saved set of real questions on the old model and the new one, and compare. A model that scores better on a benchmark can still answer your questions differently. Our post on how to know your agent still works explains how to build that question set. Remember the second gate too: for Copilot Studio, external models also have to be allowed in the Power Platform admin centre.

What we would tell a firm

1. Find out whether Anthropic is enabled in your tenant, and who decided.
2. Look at the OpenAI setting as well, and check it against what you have told clients.
3. Leave the picker on Auto for most work. Name the two or three jobs where someone should choose a model on purpose.
4. Set spending limits for Cowork before the new models make bigger tasks tempting.
5. Test any working agent against saved questions before changing its model.

The bottom line

The headline is four new models. The useful part for a professional services firm is that the gap between the everyday models and the frontier ones is narrowing, and the decisions around them have not changed: who can use which provider, where the data goes and who pays for the heavy work.

Choosing the model behind each agent, and setting the controls around it, is part of how we design and build Copilot Studio agents. If you want a straight answer on what your tenant allows today, book a free 30-minute call.

Sources checked

Last checked: 3 October 2026.

• Microsoft Tech Community, “More Models, One Copilot”, 22 September 2026
• Microsoft Tech Community, “Available today: OpenAI’s GPT-6.1 Sol and Claude Sonnet 5.5 in Microsoft Copilot”, 30 September 2026
• Microsoft Tech Community, “Evolution of the Copilot pricing model”, 25 September 2026
• Anthropic, “Introducing Claude Opus 5.5”
• OpenAI, “Introducing GPT-6 Sol and Luna”
• Microsoft Learn, “Choose a model for Copilot Cowork”
• Microsoft Learn, “Select a primary AI model for your agent”, Copilot Studio
• Microsoft Learn, “Billing rates and management”, Copilot Studio
• Microsoft Learn, “Anthropic models in Microsoft Online Services”
• Microsoft Learn, “OpenAI as a subprocessor in Microsoft Online Services”