← All tool briefs

Tool brief · July 13, 2026

GPT-5.6 in Microsoft 365 Copilot: what changes for finance

FinanceFor Finance

The tool

GPT-5.6 in Microsoft 365 Copilot

Visit GPT-5.6 in Microsoft 365 Copilot

What it is

OpenAI announced GPT-5.6, which will become the new preferred model in Microsoft 365 Copilot—in Word, Excel, PowerPoint, Chat and Cowork. You don't install anything. If your org already licenses Copilot, the model powering the sidebar in Excel and Word quietly changes underneath you.

Microsoft has switched Microsoft 365 Copilot's default AI model to OpenAI's GPT-5.6, effective July 9, 2026. The upgrade brings longer context handling, improved data analysis, and better reasoning to Word, Excel, PowerPoint, and Cowork.

The next-work-session test

Concrete scenario: it's day three of the month-end close. You have a variance workbook — actuals vs. budget by cost center, with a footnote sheet explaining reclasses. You need a commentary draft for the CFO deck by end of day.

Under the old Copilot, you'd ask for variance commentary and get generic bullets that missed the reclass sheet entirely. The vendor claim now is different: in Excel, GPT-5.6 helps Copilot reason through more complex analysis and move from task to outcome with less manual assembly. Translated into finance terms — that's the pitch for cross-sheet reasoning on a variance pack. Whether it actually reads your reclass footnotes correctly is something you'll only know by testing on a workbook where you already know the answer.

Pricing

The model swap itself is free — you get it through your existing Copilot license. The license is the cost.

The enterprise add-on is $30 per user per month on an annual term. For organizations up to 300 users, Microsoft 365 Copilot Business is $18 per user per month promotionally (standard price $21) through December 31, 2026. Confirm your quote against Microsoft's own pricing page — enterprise agreements and regional pricing vary, and the promotional tier has an expiry attached.

Note: there's no separate "GPT-5.6 tier." You don't pay OpenAI directly for this.

What we'd actually use it for

Honest, narrow list for a close/audit/treasury seat:

  • First-draft variance commentary on a workbook you already understand. Faster than typing, still requires you to check every number claim.
  • Formula scaffolding — nested IFs, SUMIFS across sheets, index/match patterns you can read and audit. Excel's Copilot should deliver more sophisticated formula suggestions—think correctly nested IF statements, VLOOKUPs with multiple criteria, and even basic macro generation — that's a vendor-adjacent claim, treat it as a starting point.
  • PBC list drafting for the auditors — asking Copilot Chat to turn a prior year's request list into this year's, with your context.
  • Reformatting bank confirmation letters and covenant compliance memos in Word from a template plus this quarter's numbers.

Limits

The model change doesn't fix the boring, real problems.

  • You still can't point it at your ERP. Copilot sees what's in the Microsoft 365 tenant — SharePoint, OneDrive, Outlook, the open workbook. Your close data likely lives in NetSuite, Workday, SAP, or a data warehouse. Bridging that gap is still an integration project.
  • Numbers still need tie-out. Financial analysis, external communications, legal language, personnel decisions, security guidance, and executive reporting require stronger review regardless of which model generated the draft. That's the vendor's own posture — take it seriously for anything that touches the audit file.
  • Reasoning ≠ correctness. The pitch is fewer hallucinated columns and better cross-sheet logic — GPT-5.6's enhanced reasoning capabilities let it parse complex spreadsheet structures, suggest formulas that actually work across multiple sheets, and generate PivotTable recommendations without hallucinating phantom columns — but that's the pitch, not an audit-quality guarantee. Assume errors until you've stress-tested it on your workbooks.
  • Admins own the rollout. Users don't need to take any action, but admins should review Copilot analytics to note any changes in behavior. Silent model swaps mid-close are exactly the kind of change control your SOX process is going to ask about.
  • No sign the model itself has been finance-tuned — it's optimized for knowledge work broadly, not for GAAP or IFRS specifics.

Try it if

  • Your team already has Copilot licenses and you've been underwhelmed by earlier drafts.
  • You spend real hours writing MD&A-style commentary or audit PBC responses from workbooks that already sit in SharePoint.
  • You have a controls process that can catch AI errors before they leave the department.

Skip it if

  • Your source of truth is a GL or consolidation tool that Copilot can't reach — you'll spend more time exporting to Excel than saving.
  • You're pre-IPO or in a heavily regulated close and can't yet document how AI-assisted content gets reviewed.
  • You were hoping this changes the Excel calc engine itself. It doesn't — GPT-5.6 sits in the sidebar. Your formulas still run in Excel.

---

This content is for informational purposes only and is not financial, investment, or accounting advice. Verify outputs against authoritative sources before use.

Source: openai.com

More for Finance professionals →

Get the next one in your inbox

One daily brief. Every story gets a hype verdict.

No spam. Unsubscribe anytime.

No sponsored verdicts · We have no paid relationship with featured vendors