Microsoft Copilot Cowork Is Now GA — And It Costs Up to $230 Per Task. Here's What Every Ops Team Needs to Know

Microsoft Copilot Cowork went GA on June 16, 2026. Usage-based billing kicked in July 1 at $0.01 per Copilot Credit. Heavy tasks can hit $230+. More than half the Fortune 500 used it during preview. Now the meter is running, and ops teams who don't set limits are going to get a nasty bill.

Microsoft dropped Copilot Cowork into general availability on June 16, and the reception has been sharply divided. On one side, you've got teams doing things that simply weren't possible before: comparing four thousand files across product versions in hours, generating dependency flow charts after every batch-job edit, turning a week of pipeline review into a single morning. On the other, you've got a pricing model that charges per task, and those tasks are already hitting $230 a pop.

Let's be clear about what Cowork actually is, because the name doesn't help. It's not the Copilot sidebar you're used to. Cowork is an agentic layer that executes long-running, multi-step tasks across Microsoft 365. You describe an outcome, and Cowork plans the steps, reasons across your emails, files, meetings, and chats through Microsoft's Work IQ context engine, and returns a finished deliverable. It keeps running even when your laptop is closed.

That's powerful. It's also expensive, and the billing model deserves every bit of scrutiny it's getting.

What does a Cowork task actually cost?

Microsoft bills Cowork in Copilot Credits, priced at $0.01 per credit on pay-as-you-go. Four inputs determine the cost of each task: model use, context retrieval, tool calls, and runtime. During the Frontier preview, Microsoft observed three common task shapes:

• **Light tasks** (~100-300 credits / $1-$3): a few sources, light reasoning, one short deliverable. Think Monday status update.
• **Medium tasks** (~400-700 credits / $4-$7): many sources, structured reasoning, two or more outputs. A meeting-prep doc plus Excel plus slide.
• **Heavy tasks** (700+ credits / $7+): broad aggregation, deep reasoning, many outputs. Classifying six months of usage data into a leadership report.

Those numbers are based on Anthropic Opus 4.8. The $230 figure comes from complex, multi-model heavy tasks that chain Opus 4.8 with GPT-5.5 (available to Frontier customers) and pull context from across the entire organisation. A single deep research brief with citation mapping and an executive dashboard can burn through 23,000 credits before you blink.

Microsoft offers two payment tracks: Pay As You Go at $0.01/credit, and Prepaid Credits (P3) at a discounted rate for those who commit to volume upfront. The sensible play for most teams is to start on PAYG, build a usage baseline over 4-6 weeks, then switch to P3 once you know what your actual consumption looks like. But a lot of teams won't do that. And that's the problem.

Is the work worth the credits?

This is the question every ops team should be asking before they enable Cowork, not after the first invoice lands. Some tasks justify the spend:

• **One-off deep analysis**: when the alternative is two people spending a week manually collating data, $200 starts looking cheap.
• **High-stakes pipeline review**: a ranked list of at-risk opportunities with exact follow-up context, produced in a morning, for $150.
• **Cross-system reconciliation**: comparing thousands of records across two systems where the manual alternative is error-prone and slow.

And then there's everything else. Running a daily standup summary through Cowork at $3/day/person is $60/month/person, which a decent no-code automation could do for a flat $30/user/month with no per-task meter. Multiply that across a 50-person ops team and you're looking at $36,000/year vs. $18,000/year for automated summaries. The no-code version doesn't get more expensive when you run it more often.

Here's a heuristic: if the task involves creative reasoning across unstructured data that changes shape every time, Cowork probably earns its keep. If the task follows a pattern (same data sources, same outputs, same logic every Tuesday), building a dedicated automation in a governed no-code platform will almost certainly be cheaper and more reliable over 12 months.

The Cost Management Dashboard and why you need it now

Microsoft's admin centre now includes a Cost Management dashboard that's worth knowing about before you enable Cowork at all. It gives you:

• Real-time credit consumption tracking and remaining capacity
• Per-user, per-group, and per-service breakdowns of spending
• Budget thresholds with alerts
• Hard caps that prevent overspend

Most importantly: **Cowork is off by default.** No task runs until an admin explicitly enables usage-based billing and activates a spending policy. If you haven't configured billing, your users literally cannot use Cowork. That's the good news. The bad news is that once it's on, without per-user caps, one person running heavy tasks all day can consume credits meant for the entire organisation. Set those caps before you enable anything.

Frontier customers who used Cowork between March 30 and June 16 got a grace period until July 1. That deadline has passed. If billing wasn't configured, access was suspended automatically.

What multi-model routing means for no-code builders

At GA, Cowork runs on Anthropic's Opus 4.8 and Sonnet 4.6. Frontier customers can also use GPT-5.5, and Microsoft's own cost-optimised Cowork 1 model is coming soon. The model picker lets you choose which model handles which task: lighter models for routine work, frontier models for complex reasoning.

For no-code builders, this matters in two ways. First, it confirms that the AI layer underneath productivity tools is becoming a commodity. You're not locked into one provider's model, and cost optimisation is increasingly about routing the right task to the right model. Second, it means the capability gap between what an enterprise AI assistant can do and what a well-built no-code automation can do is narrowing for structured work. Cowork is impressive at unstructured reasoning. For repeatable, process-driven automation with predictable costs, it's overkill and overpriced.

The elephant in the room: consumption-based pricing

There's a pattern forming across Microsoft's AI portfolio. GitHub Copilot moved to usage-based billing on June 1. Copilot Cowork followed on June 16. Work IQ API now bills per call. The message is clear: Microsoft is betting that AI features will generate enough value that organisations will accept variable, unpredictable billing.

The problem is that ops teams don't work that way. Ops teams have budgets. They have quarterly forecasts. They need to know what their tooling costs with reasonable certainty, and "somewhere between $3 and $230 per task, depending" is not a number you can put in a budget spreadsheet.

Governed no-code platforms take the opposite approach: flat per-user pricing with no usage metering. You build the automation once, it runs as many times as you need, and the cost is predictable from day one. For the 80% of automation work that's process-driven rather than reasoning-intensive, that's not just cheaper. It's saner.

The takeaway

Copilot Cowork is impressive technology. The ability to hand off complex, multi-step tasks to an agent that works across your entire Microsoft 365 environment and returns finished deliverables is a real capability advance. But the pricing model is designed for organisations that treat AI as a strategic investment with flexible budget, not for ops teams that need to automate processes at scale with predictable costs.

Before you switch on Cowork, do three things: enable the Cost Management dashboard and set per-user caps, run a month of PAYG to establish your actual usage baseline, and audit your team's automation needs. For the creative, unstructured reasoning tasks, Cowork will earn its keep. For everything else, a governed no-code platform gives you the same outcomes at a fixed price, and nobody gets a surprise $230 line item on the invoice.