Microsoft announced on July 9, 2026, that GPT-5.6 is now the preferred AI model across five Copilot surfaces in Microsoft 365, including Word, Excel, PowerPoint, Copilot Chat, and Copilot Cowork. However, the designation as "preferred" does not mean exclusive or superior across all tasks. Anthropic's Claude models still lead GPT-5.6 on several head-to-head benchmarks, including specialized coding tasks, meaning teams will need to choose their model deliberately based on the work at hand.

The rollout happened the same day GPT-5.6 reached general availability through ChatGPT, Codex, and the OpenAI API. Microsoft framed the integration as strengthening "agentic, multi-step work," positioning the model to handle complex, multi-step tasks that span across files and tools rather than delivering faster single-turn answers. The timing reflects a broader shift in how AI assistants are being deployed in productivity software, moving from simple question-answering to handling entire workflows.

What Does "Preferred" Actually Mean for Microsoft 365 Users?

The term "preferred model" carries a specific but limited meaning. It signals that Microsoft will auto-select GPT-5.6 when it judges the model best suited for a task, but users can also manually choose it from the model selector when available. This is not an exclusive arrangement. Copilot Cowork, for example, remains explicitly multi-model, with users able to select from Claude Sonnet 5, Claude Opus 4.8, GPT 5.5 (Frontier), Claude Fable 5, and a Sonnet plus Opus Advisor mode.

The rollout is also phased, not instant. Microsoft explicitly stated that availability may vary by region and tenant configuration, pointing administrators to the Microsoft 365 Roadmap and Copilot release notes. Organizations should plan around a rolling window rather than expecting immediate deployment across all users.

How to Prepare Your Organization for the GPT-5.6 Rollout

• Review Model Selection Settings: Confirm that your Copilot Cowork configuration allows users to select models manually, ensuring teams can route high-stakes work to Claude when benchmarks show it performs better for their specific task type.
• Check Governance and Data Handling Policies: GPT-5.6 slots into the existing model-picker and consent system, so Cowork's no-retention default, retention banner behavior, and the Anthropic-family toggle all remain unchanged from current configurations.
• Plan for Phased Adoption: Build a repeatable verify-and-adopt habit rather than treating each model refresh as an emergency, since Microsoft has shown it refreshes preferred models roughly every seven months, as it did with GPT-5.2 in December 2025.

Where Claude Still Outperforms GPT-5.6

On OpenAI's own general availability evaluation tables, Claude models maintain a measurable lead on several benchmarks. Most notably, on SWE-Bench Pro, a specialized coding benchmark, Claude Fable 5 scored 80 compared to GPT-5.6 Sol's 64.6. This gap is significant for teams relying on AI for software engineering tasks, code review, or complex debugging workflows. The difference suggests that routing coding-heavy work to Claude rather than accepting the auto-selected GPT-5.6 could yield better results.

Governance and data handling remain unchanged across the transition. The existing consent system, retention policies, and admin settings layout all carry over untouched, meaning IT teams do not need to reconfigure security or compliance controls.

What Changes in Each Microsoft 365 App?

Microsoft and OpenAI described GPT-5.6's expected impact across each application. In Word, the model is positioned to help users draft, edit, and refine documents with fewer rounds of prompting and stronger organization. In Excel, the claim centers on deeper analysis while using tokens more efficiently, moving faster from raw data to insights without manual assembly. PowerPoint should see richer presentation drafts with stronger slide content and better visual balance.

Copilot Chat is positioned for reasoning through more complex requests, such as comparing options, structuring plans, troubleshooting problems, and working through ambiguity. The biggest framing shift appears in Copilot Cowork, where GPT-5.6 is expected to strengthen agentic work by carrying complex multi-step tasks from instruction to completed results, paired with Microsoft's Work IQ grounding layer.

"Microsoft 365 is where millions of people write, analyze, create, and collaborate every day. By bringing GPT-5.6 to Microsoft 365 Copilot through the OpenAI API, we're helping organizations get more useful work from every token, and more value from AI in the tools they already use," said Nikunj Handa, Head of API Product at OpenAI.

Nikunj Handa, Head of API Product, OpenAI

For business teams, PowerPoint and Copilot Cowork warrant the closest attention. Deck production is one of the most time-intensive recurring tasks in marketing and client-services work, so improvements in slide generation and visual balance could meaningfully reduce production time. Cowork's "finished deliverable" promise is where the agentic capabilities get tested in real workflows.

Pricing and Capability Tiers for GPT-5.6

OpenAI released GPT-5.6 in three pricing tiers at general availability. The flagship tier costs $5 per million input tokens and $30 per million output tokens, with cached-input discounts available. This is likely the engine behind Copilot's most complex reasoning tasks, though Microsoft has not published per-app tier mapping. The balanced tier runs $2.50 input and $15 output per million tokens, while the high-volume tier costs $1 input and $6 output per million tokens. OpenAI positioned these three tiers as durable capability levels rather than temporary pricing, making API budgeting more predictable for organizations.

The context window size and general availability rate limits for GPT-5.6 were unpublished at the time of the announcement, so organizations should verify those specifications before finalizing cost projections.

This rollout reflects a broader pattern. Microsoft ran the same preferred-model refresh for GPT-5.2 in December 2025, roughly seven months before this announcement. Rather than treating each refresh as a one-off emergency, organizations should establish a repeatable process for evaluating and adopting new preferred models as they arrive on a predictable cadence.