Claude Opus 4.8 is now in Microsoft 365 Copilot — here's what changes for you
Anthropic's Claude Opus 4.8 arrives in Microsoft 365 Copilot on day one, expanding to Copilot Chat, Excel, PowerPoint, and Copilot Studio with stronger reasoning and Work IQ grounding.
Anthropic released Claude Opus 4.8 on 28 May 2026, and Microsoft had it live inside Microsoft 365 Copilot the same day. That kind of turnaround is worth noting on its own, but the more significant story is where it lands and what it actually does differently.
Where Opus 4.8 shows up in Microsoft 365
Claude Opus 4.8 is available today in the model selector inside Copilot Cowork (Frontier). It is rolling out to Microsoft 365 Copilot licensed users in:
- Copilot Chat
- Excel
- PowerPoint
- Copilot Studio (early release cycle environments)
In Copilot Cowork specifically, Opus 4.7 has been replaced by Opus 4.8 outright. Broader availability also extends to Researcher, editing within Office apps, GitHub Copilot, and Microsoft Foundry.
This is not a separate product you open in a new tab. When your admin has enabled Anthropic models for your tenant, Claude appears directly in the Copilot model selector alongside other available options.
What Opus 4.8 adds over 4.7
Opus 4.8 is built for complex, multi-step work. The improvements over 4.7 centre on three areas: better tool selection, closer adherence to instructions, and stronger follow-through across multi-turn workflows.
The benchmark numbers give a sense of the jump. On SWE-bench Pro (a coding reliability test), Opus 4.8 scores 69.2%, up from 64.3% on Opus 4.7. The maths improvement is more striking: 96.7% on USAMO 2026, compared with 69.3% for Opus 4.7. Anthropic also reports that Opus 4.8 fails to surface important events to users only 3.7% of the time, and scores 0% on uncritically reporting flawed results — the first Claude model to reach that threshold.
For everyday use, Opus 4.8 is also stronger at drafting documents, analysing data, and building presentations. When paired with Work IQ grounding (more on that below), those outputs draw on your organisation’s actual context rather than generic responses.
Fast mode for Opus 4.8 is now three times cheaper than it was for previous models, while still running at around 2.5x the speed of standard mode. That makes it more practical for higher-volume or time-sensitive tasks.
Work IQ grounding: why this matters
Work IQ is the intelligence layer underneath Microsoft 365 Copilot that connects signals from Outlook, Teams, Excel, files, and meetings to give Copilot shared organisational context. When Opus 4.8 runs inside Copilot Cowork, it is automatically grounded in that context. The model can draw on your emails, meeting history, and documents to produce outputs that are relevant to your actual work rather than starting cold.
Work IQ is designed to respect existing user permissions, security group assignments, sensitivity labels, and DLP policies. You are not opening up a broader data surface by using Claude inside Copilot. The same governance boundaries that apply to the rest of Copilot apply here.
The EU, EFTA, and UK situation
For most commercial tenants outside the EU, EFTA, and UK, Anthropic models are turned on by default. If your organisation is already opted in, there is nothing to do.
For organisations in the EU, EFTA, and UK, Claude is off by default. Admins need to opt in via the Microsoft 365 Admin Center. Government and sovereign clouds are excluded for now, with updates expected.
This is worth flagging for IT teams in those regions. If your users are asking why they cannot see Claude in the model selector, this is likely the reason.
If you build agents in Copilot Studio
One important operational note: agents you have built in Copilot Studio will not be automatically updated to use Opus 4.8. Older model versions remain in place for existing agents, giving you time to test the new model and confirm that your agents behave as expected before switching. This is deliberate, and it is the right call. Model upgrades can change output behaviour in ways that matter for scheduled prompts or structured outputs, so the manual upgrade path gives you control.
Bear in mind that older models do eventually get retired in Office apps and Copilot Chat, so this is not a permanent reprieve. It is a window to test properly.
The bigger picture
A few months ago, Copilot was effectively a single-model service built around OpenAI’s GPT family. That is no longer the case. Microsoft is now treating model selection as a deliberate product and governance decision, letting organisations match workloads to the model characteristics that suit them best, whether that is reasoning depth, cost, latency, or data residency requirements.
Opus 4.8 arriving on day one of its public release shows how that multi-model approach works in practice. The model is available, it is grounded in your organisational data through Work IQ, and it runs inside the same security and compliance boundaries you already have in place.
If you are a Microsoft 365 Copilot user in a tenant that has Claude enabled, you will start seeing Opus 4.8 in the model selector as the rollout completes. If you are an admin in the EU, EFTA, or UK and want to make it available, the opt-in is in the Microsoft 365 Admin Center.