Hero image for Copilot Pay-As-You-Go Pricing: Ignite 2026 Verdict
By AI Tool Briefing Team

Copilot Pay-As-You-Go Pricing: Ignite 2026 Verdict


Ignite opened November 17 in San Francisco, and Satya Nadella spent the keynote selling Copilot as the work operating system. Constellation Research summed up the pitch that way. The money side of that pitch is Microsoft’s own explanation of its new Copilot pay-as-you-go pricing model: a fixed per-user seat for “everyday” AI, and metered usage for “advanced” AI.

Here’s the part buyers should sit with. The model isn’t new this week. Microsoft published it on September 25. Ignite was the stage, not the announcement. And the keynote didn’t hand over what a finance team actually needs: a rate card, an Agent 365 price list, or dates.

Quick Summary: Copilot Pricing After Ignite 2026

DetailInfo
EventMicrosoft Ignite 2026, November 17-20
ModelPer-user license for everyday AI, plus usage-based billing (UBB) in Copilot Credits for advanced AI
Seat priceStill cited at $30/user/month for Microsoft 365 Copilot (Microsoft’s pricing post gives no dollar figure)
Credit price$0.01 per credit pay-as-you-go, per Futurum
Agent 365$15/user/month at May GA; no new price list surfaced at Ignite
Microsoft’s cost chartVendor-modeled, unverified
Official sourceMicrosoft Copilot blog

Bottom line: Copilot is becoming a seat fee plus a meter. The seat is predictable. The meter isn’t, and nobody has published how fast it runs. Don’t renew on the strength of a keynote slide.


What Microsoft Actually Announced

Start with the confirmed material. Microsoft’s pricing post, dated September 25 and updated September 28, splits Copilot into two buckets.

Everyday AI stays on the per-user license. The post says it includes OpenAI’s GPT-5.6 and Anthropic’s Sonnet 5, with Opus 5 included “with limits.” An Auto setting routes each request to a model and reasoning level. Heavy users who hit a limit get a warning, then can drop to Auto at no extra cost or move to credits.

Advanced AI is metered. That means long-running agents and frontier models: Copilot Cowork, Code and Autopilot are the named examples. These run on usage-based billing through Copilot Credits, and the per-user license is required to access them. So you can’t skip the seat.

Futurum’s analysis adds two details: pay-as-you-go credits are $0.01 each, with pre-purchase discounts of up to 20%, and enterprise usage billing stays off until an admin creates a spending policy in the Microsoft 365 admin center. That last one is a genuinely useful guardrail. Microsoft also maintains a Copilot Credits licensing guide for the mechanics.

Notice what’s missing from Microsoft’s own post: it gives no price per credit and no effective date. The $0.01 figure comes from secondary coverage of the licensing guide. Confirm it against your agreement.

What the Keynote Left Out

According to Redmond’s coverage of the keynote, the Ignite announcements were product-shaped: Copilot Home combining Chat and Cowork, Copilot Managed Runtime going generally available, Autopilot rolling out to Frontier customers, Copilot Workspaces for shared projects, and Copilot Code expected “in the coming weeks.” That article carries no pricing for any of it.

Some reports say the keynote skipped a Copilot Credits rate card, an Agent 365 price list and dates for the transparency standards, and that Workspaces and Autopilot pricing is unpublished. No linkable write-up backs that up, so treat it as unconfirmed. What the cited keynote coverage does show: none of those numbers appear.

That gap matters because Autopilot and Workspaces are exactly the products most likely to burn credits. Autonomous, long-running work is the whole point of a meter.

What Is Microsoft’s Copilot vs ChatGPT vs Claude Cost Chart?

Microsoft’s chart is an internal cost analysis, not an independent benchmark. The pricing post says it compared Copilot with Claude and ChatGPT across 40 prompts, each run 10 times, and that costs exclude discounts. Microsoft’s chart reportedly shows much higher costs for both rivals than for a $30 “everyday” Copilot user, but this post could not verify those rival figures.

In Microsoft’s post the results sit in an image, not text, and no outlet reports them. Treat any specific rival dollar figure you see as unverified.

Futurum’s caveats are the useful ones:

  1. It’s vendor-modeled. Microsoft built the test and picked the profiles.
  2. It uses list and API pricing. No negotiated discounts, which is how most enterprises actually buy.
  3. It’s not like-for-like. Copilot Cowork at variable rates was compared against competitors’ public pricing.
  4. Microsoft is both reseller and rival. Copilot runs on OpenAI and Anthropic models, so Microsoft is benchmarking against the labs it resells.

There’s also a framing trick worth naming. Microsoft’s post takes a swipe at “some vendors” that meter everyday AI, arguing that metering makes companies ration access. Fair argument. But a chart that prices a flat $30 against rivals’ usage rates only wins if your people stay inside the fair-use limits. Microsoft decides where that line sits.

For the real sticker prices of the rivals, see our AI pricing comparison.

Copilot Pay-As-You-Go Pricing: The Seat Isn’t the Whole Bill

Microsoft 365 Copilot is still cited at $30 per user per month. We covered the earlier tier split, $18 Copilot Business versus $30, in our Copilot pricing breakdown. That post predates the metering shift, so read it as the seat-fee half of the story only.

Agent 365 adds a second license. It went GA in May at $15 per user per month standalone, and our Agent 365 buyer’s guide walks through who needs a seat. One more wrinkle: sources disagree on its scale. CryptoBriefing reported 40 million agents registered within two months of launch. That number isn’t Microsoft-confirmed in the cited coverage, so don’t quote it in a business case.

If this billing shape looks familiar, it should. GitHub went through the same move, and we dug into it in GitHub Copilot’s token billing. Seats get you in the door. Consumption is where the revenue grows.

The Cheaper Models and the Maia 300 Chip

Microsoft’s other Ignite message was cost: it wants to run more of Copilot on its own models.

Axios reported six new Microsoft models announced at Ignite, built for specific tasks to reach comparable performance at lower cost, plus licensed model weights so customers can customize them with their own data (deployment must stay in Azure). Thurrott counts seven, so the exact lineup is inconsistent across outlets. Finimize reports a cheaper coding model, MAI Code 2, and a security model in limited preview. For the earlier MAI launch and why Microsoft is hedging against OpenAI, see our MAI models explainer.

Then the silicon. Microsoft detailed Maia 300 in an October 12 Azure post: eight HBM4 stacks, 288 GB and 18 TB/s per accelerator, a 96-accelerator rack, and up to 7x the rack-scale token throughput of Maia 200.

The headline claim is about 40% more tokens per megawatt than leading GPU systems. Note the unit. Some summaries say per watt; Analytics India Magazine says per megawatt. And as Let’s Data Science points out, Microsoft doesn’t say which GPUs it compared against. The 40% line doesn’t appear in the Azure architecture post itself. NeoTeo reports production deployment in 2027.

So: a chip that arrives next year, with an unnamed comparison baseline, can’t be your reason to expect lower Copilot bills in your next renewal. Maybe it lowers Microsoft’s costs. Whether that reaches you is a negotiation, not a spec sheet.

Our Take

I think the two-tier structure is reasonable. Flat pricing for chat and summaries, metering for agents that run for hours: that’s roughly how the underlying compute behaves. My objection is to the sequencing. Microsoft is asking you to sign a seat agreement now and trust the meter later.

Futurum’s point lands hard: finance teams can cap spend, but they can’t forecast what an agentic task costs before it runs. Credits track compute and runtime, not outcomes. An agent that grinds for hours on a bad result costs the same as one that nails it. That’s the thing no keynote chart addresses.

The cost chart is marketing. Useful marketing, since it tells you which comparison Microsoft wants you making, but marketing.

What to Ask Your Reseller Before Renewal

  1. What is the credit rate in my agreement? Get it in writing, not the $0.01 from an analyst article.
  2. What triggers metering? Ask for a list of which features sit in the seat versus UBB today, and which are scheduled to move.
  3. Can we hold UBB off? Confirm it stays disabled until an admin sets a spending policy, and get the cap mechanics.
  4. What is the Agent 365 price for our case? Specifically who counts as a user.
  5. What are the fair-use limits on Opus 5 and Auto? And can Microsoft change them mid-term?
  6. Will you commit to migration terms? If a capability moves from metered to seat-included, or the reverse, put the rules in the contract.
  7. Can you show the chart’s inputs? Prompts, profiles, discounts excluded. Then rerun it with your own usage.

Frequently Asked Questions

Is Microsoft 365 Copilot still $30 per user per month? That’s the figure still cited for the add-on. Microsoft’s September pricing post gives no dollar amount, so confirm it on your quote. Usage-based charges for advanced features come on top.

What are Copilot Credits? The metered currency for advanced AI such as long-running agents. Per Futurum, pay-as-you-go is $0.01 per credit with up to 20% off on pre-purchase.

Do I still need the per-user license? Yes. Microsoft’s post says the license is required to access usage-based capabilities.

Is Microsoft’s cost comparison against ChatGPT and Claude reliable? Not as published. It comes from Microsoft’s own modeling, excludes discounts, and the rival figures can’t be checked from the source image.

Did Ignite announce Agent 365 pricing? Nothing in the cited keynote coverage. The last confirmed price is $15 per user per month from May.

When does Maia 300 reach customers? Reports point to production deployment in 2027. The 40% efficiency figure is Microsoft’s claim and the comparison baseline is unnamed.


Last updated: November 19, 2026. Sources: Microsoft Copilot blog, Microsoft Copilot Credits guide, Futurum, Redmond Channel Partner, Constellation Research, CryptoBriefing, Axios, Thurrott, Finimize, Azure blog, Analytics India Magazine, Let’s Data Science, NeoTeo. Related reading: Microsoft Build 2026 and Project Polaris, Copilot pricing after the Office changes, Agent 365 for enterprise buyers.