Microsoft is moving Copilot Cowork, its enterprise AI agent for Microsoft 365 work, to usage-based pricing as it reaches broader availability in June 2026, while reportedly evaluating a Microsoft-hosted DeepSeek model as a lower-cost option for customers running long, compute-heavy tasks. The pricing change is not a bookkeeping tweak; it is Microsoft admitting that autonomous office agents do not fit comfortably inside the old per-seat software model. The DeepSeek angle makes the move more combustible, because the cheapest path to scalable enterprise AI may also be the path that drags geopolitical risk directly into the Microsoft 365 admin center.
Copilot Cowork is supposed to be the version of Copilot that stops merely suggesting edits and starts doing work. It can plan tasks, reason across Microsoft 365 data, draft documents, prepare presentations, and keep chipping away at multi-step assignments after the user has moved on. That is the sales pitch behind the agentic turn: AI not as a chat box, but as a colleague with a queue.
The problem is that a colleague draws a salary, and an agent burns tokens. A conventional Microsoft 365 license maps neatly to a user, a mailbox, a set of apps, and a predictable monthly bill. Cowork is different because the expensive part happens after the user clicks “go”: model calls, tool calls, retries, context expansion, document grounding, web lookups, and repeated reasoning loops.
That is why usage-based pricing was probably inevitable. If a user asks Copilot to summarize a page, Microsoft can absorb or average out the cost. If a department tells Cowork to monitor projects, rewrite decks, generate reports, compare data, and run workflows for hours, the flat-rate bundle starts looking less like SaaS and more like an all-you-can-eat buffet being raided by forklifts.
Microsoft’s challenge is not simply that AI is expensive. It is that the most compelling AI demos are the least compatible with predictable subscription economics. The more Cowork succeeds at becoming useful, the more it threatens the tidy licensing model that made Microsoft rich.
That worked for chat-style Copilot because ordinary use is bursty. A user asks for a rewrite, a summary, or a meeting recap, then stops. The cost can be estimated across a large base of subscribers, and Microsoft can decide how much margin it is willing to sacrifice to win adoption.
Agentic AI breaks that bargain. The user does not just ask; the user delegates. Delegation means the system may call a model repeatedly, inspect intermediate results, invoke tools, gather more context, and attempt the task again when the output is not good enough. The cost is no longer tied to the number of humans with seats. It is tied to how much work the AI is allowed to attempt.
This is the same pressure now visible around developer agents. GitHub Copilot’s shift toward usage-based billing exposed the awkward truth that coding assistants can become very different products depending on whether they are completing a line of code or orchestrating a multi-file refactor. One is a convenience feature. The other is a compute-consuming workflow engine.
Cowork brings that dynamic into Microsoft 365. Instead of developers watching token consumption during a coding session, finance and IT teams may now have to watch AI spend across Word, Excel, PowerPoint, Outlook, SharePoint, Teams, and connected business systems. The meter has arrived in the office suite.
That is the rational engineering story. It is also incomplete.
DeepSeek is not just another model family in a benchmark table. Since its rise, it has been treated by many Western governments and security analysts as a test case for the risks of Chinese AI systems: data exposure, censorship behavior, national security concerns, and dependence on technology developed under a rival legal and political regime. Even if Microsoft hosts the model inside Azure, keeps customer data within Microsoft-controlled infrastructure, and subjects the service to Azure compliance and residency controls, the name on the model will still matter to procurement teams, regulators, and politicians.
Microsoft knows this. That is why the reported plan emphasizes optionality and Azure hosting. The argument will be that customers are not sending data to DeepSeek’s public service, not using DeepSeek’s consumer app, and not stepping outside Microsoft’s enterprise security boundary. In Microsoft’s framing, this would be a model choice inside a governed cloud platform, not a data handoff to a Chinese company.
That distinction is technically important. It may also be politically insufficient. Enterprise security buyers do not evaluate risk only by packet flow diagrams. They evaluate vendor exposure, supply-chain narratives, regulatory climate, board-level optics, and whether a decision will still look defensible after the next congressional hearing or government memo.
For many commercial customers, that may be enough. A bank, manufacturer, or software company might decide that a cheaper Azure-hosted model is acceptable for low-sensitivity work, especially if Microsoft provides clear controls for model selection, logging, and policy enforcement. A tenant admin could allow less expensive models for summarization and drafting while restricting regulated data to approved models.
But the hard cases will define the controversy. Government agencies, defense contractors, critical infrastructure operators, and heavily regulated enterprises may not care that the workload stays in Azure if the underlying model lineage is politically radioactive. Their question will not be “Where is the prompt processed?” but “What model behavior, training provenance, and legal exposure are we accepting?”
This is where Microsoft’s multi-model strategy becomes both powerful and messy. Customers want choice because no single model is best at every task or price point. They also want someone else to absorb the complexity of deciding which models are safe, compliant, performant, and cost-effective. Microsoft wants to be that abstraction layer, but abstraction does not eliminate accountability.
If Cowork quietly routes work across different models, admins will demand visibility. If Microsoft exposes model choice directly, admins will need policy tools. Either way, model governance becomes part of Microsoft 365 administration. The AI era is turning the tenant admin into a model-risk manager.
This will be uncomfortable for organizations that treated Microsoft 365 as predictable overhead. A user-based license can be negotiated, allocated, and forecast with relative confidence. A usage-based agent can surprise you, especially if a small group of power users discovers workflows that run constantly or generate repeated high-context requests.
The likely result is a new internal politics of AI access. Executives will want Cowork enabled because the demos look transformative. Finance will want limits because variable bills are hard to explain. Security will want model restrictions. Legal will want assurances about data use. End users will want the system to work without being nagged about credits.
Admins will be caught in the middle. They will need tools that answer basic questions Microsoft has not historically had to expose in Office: Which users are consuming the most AI compute? Which workflows are expensive? Which models are being used? Which data sources are being grounded? Which tasks failed and retried? Which business unit owns the bill?
The winners will be organizations that treat Cowork like a new class of cloud workload rather than a smarter Clippy. The losers will be organizations that enable it broadly, celebrate early adoption, and only later discover that autonomous office work has an autonomous cost structure.
A multi-model strategy gives Microsoft leverage. It can route cheaper tasks to cheaper models, reserve expensive systems for complex reasoning, and pressure model suppliers on price. It can also reassure customers that Copilot is not simply a wrapper around one vendor’s roadmap.
This is not a repudiation of OpenAI or Anthropic. It is a recognition that enterprise AI will look more like cloud infrastructure than a single magical assistant. Different workloads will demand different latency, cost, accuracy, privacy, and compliance characteristics. The model layer will become a portfolio.
That portfolio approach is exactly what Microsoft has been building through Azure AI Foundry, Copilot Studio, Microsoft 365 Copilot, and its broader agent platform. The company wants to be the broker, host, policy engine, billing system, and user interface for workplace AI. In that role, Microsoft benefits when models become interchangeable components.
But interchangeability cuts both ways. If customers learn that a cheaper open model is “good enough” for a large share of office work, Microsoft’s premium AI story becomes harder to price. If customers decide model origin matters more than cost, Microsoft’s bargain option may remain confined to less sensitive workloads. The multi-model future is flexible, but it is not simple.
That is why Microsoft’s broader agent governance push matters. As AI agents become embedded in Microsoft 365, admins will need controls that look familiar from identity, compliance, and cloud management. They will need policy scopes, audit logs, approval flows, spend caps, data boundaries, and model allow-lists.
The old Office admin model was built around users and applications. The new model must account for non-human actors performing work on behalf of users across applications. That changes the risk profile. A human making a mistake in Excel is one thing. An AI agent repeatedly applying a mistaken instruction across files, emails, meetings, and connected systems is another.
Cowork’s move to usage-based pricing reinforces that governance problem because cost and control are now linked. An unrestricted agent is not only a security risk; it is a budget risk. A poorly designed workflow is not only inefficient; it is expensive. A hallucinated detour is not only embarrassing; it may consume billable compute.
For WindowsForum readers, this is the part to watch. Microsoft’s success will not depend only on whether Cowork can draft a better deck or schedule a better meeting. It will depend on whether admins can make agentic work governable at enterprise scale.
But the market will hear the subtext. Microsoft is searching for cheaper inference because agentic AI is expensive to run. It is willing to consider politically sensitive models because cost matters. And it believes Azure’s trust boundary can turn a controversial model into an enterprise-acceptable service.
That is a bold bet. It assumes customers will distinguish between using DeepSeek as a public Chinese-hosted chatbot and using a Microsoft-hosted model derived from DeepSeek technology. Some will. Others will not. In sectors where vendor risk assessments already move slowly, the phrase “Chinese AI model” may stop the conversation before Azure residency controls can rescue it.
Microsoft can reduce the blast radius by being unusually transparent. It should disclose the model identity, hosting architecture, data handling rules, evaluation results, limitations, and admin controls. It should make model selection visible rather than burying it behind a friendly Copilot label. It should let customers disable entire model families if their policies require it.
The worst version of this rollout would be ambiguity: a cheap Cowork mode with vague model language, unclear routing, and marketing-heavy assurances about trust. The best version would be boringly explicit. In enterprise AI, boring is a feature.
That means users should expect a more tiered Copilot experience over time. Basic chat, summarization, and drafting may remain bundled into existing plans. Longer-running agents, premium reasoning models, and high-volume automation will increasingly sit behind credits, meters, or enterprise controls. The more the AI does, the more the invoice will look like cloud consumption.
For administrators, this is a reminder to slow down before turning on the shiny new switch. Cowork may be genuinely useful, especially for repetitive knowledge work and cross-document tasks. But broad enablement without policy is a recipe for confusion. Usage-based pricing rewards experimentation at first and punishes unmanaged enthusiasm later.
The same caution applies to model choice. A cheaper model may be perfectly acceptable for low-risk workloads. It may be unacceptable for sensitive legal, government, defense, financial, or regulated use. The right answer will vary by organization, which means Microsoft must give admins the knobs to express that policy.
Microsoft’s Agent Dream Meets the Invoice
Copilot Cowork is supposed to be the version of Copilot that stops merely suggesting edits and starts doing work. It can plan tasks, reason across Microsoft 365 data, draft documents, prepare presentations, and keep chipping away at multi-step assignments after the user has moved on. That is the sales pitch behind the agentic turn: AI not as a chat box, but as a colleague with a queue.The problem is that a colleague draws a salary, and an agent burns tokens. A conventional Microsoft 365 license maps neatly to a user, a mailbox, a set of apps, and a predictable monthly bill. Cowork is different because the expensive part happens after the user clicks “go”: model calls, tool calls, retries, context expansion, document grounding, web lookups, and repeated reasoning loops.
That is why usage-based pricing was probably inevitable. If a user asks Copilot to summarize a page, Microsoft can absorb or average out the cost. If a department tells Cowork to monitor projects, rewrite decks, generate reports, compare data, and run workflows for hours, the flat-rate bundle starts looking less like SaaS and more like an all-you-can-eat buffet being raided by forklifts.
Microsoft’s challenge is not simply that AI is expensive. It is that the most compelling AI demos are the least compatible with predictable subscription economics. The more Cowork succeeds at becoming useful, the more it threatens the tidy licensing model that made Microsoft rich.
The Flat-Rate Copilot Era Was Always a Subsidy
Microsoft has spent the past few years trying to make Copilot feel like the natural next layer of Windows, Office, GitHub, Teams, Outlook, and Azure. That strategy required one psychological move above all others: hide the meter. Users were meant to think of AI as another productivity feature, not as a cloud workload with an invisible GPU meter spinning behind every answer.That worked for chat-style Copilot because ordinary use is bursty. A user asks for a rewrite, a summary, or a meeting recap, then stops. The cost can be estimated across a large base of subscribers, and Microsoft can decide how much margin it is willing to sacrifice to win adoption.
Agentic AI breaks that bargain. The user does not just ask; the user delegates. Delegation means the system may call a model repeatedly, inspect intermediate results, invoke tools, gather more context, and attempt the task again when the output is not good enough. The cost is no longer tied to the number of humans with seats. It is tied to how much work the AI is allowed to attempt.
This is the same pressure now visible around developer agents. GitHub Copilot’s shift toward usage-based billing exposed the awkward truth that coding assistants can become very different products depending on whether they are completing a line of code or orchestrating a multi-file refactor. One is a convenience feature. The other is a compute-consuming workflow engine.
Cowork brings that dynamic into Microsoft 365. Instead of developers watching token consumption during a coding session, finance and IT teams may now have to watch AI spend across Word, Excel, PowerPoint, Outlook, SharePoint, Teams, and connected business systems. The meter has arrived in the office suite.
DeepSeek Is the Cost-Cutting Option Microsoft Cannot Present as Merely Technical
The reported DeepSeek evaluation is easy to understand in economic terms. If Cowork’s cost problem is driven by repeated model calls, then Microsoft needs cheaper models for tasks that do not require the most expensive frontier systems. A fine-tuned DeepSeek V4 or another open model could give Microsoft a lower-cost path for routine agent work while preserving premium models for more demanding tasks.That is the rational engineering story. It is also incomplete.
DeepSeek is not just another model family in a benchmark table. Since its rise, it has been treated by many Western governments and security analysts as a test case for the risks of Chinese AI systems: data exposure, censorship behavior, national security concerns, and dependence on technology developed under a rival legal and political regime. Even if Microsoft hosts the model inside Azure, keeps customer data within Microsoft-controlled infrastructure, and subjects the service to Azure compliance and residency controls, the name on the model will still matter to procurement teams, regulators, and politicians.
Microsoft knows this. That is why the reported plan emphasizes optionality and Azure hosting. The argument will be that customers are not sending data to DeepSeek’s public service, not using DeepSeek’s consumer app, and not stepping outside Microsoft’s enterprise security boundary. In Microsoft’s framing, this would be a model choice inside a governed cloud platform, not a data handoff to a Chinese company.
That distinction is technically important. It may also be politically insufficient. Enterprise security buyers do not evaluate risk only by packet flow diagrams. They evaluate vendor exposure, supply-chain narratives, regulatory climate, board-level optics, and whether a decision will still look defensible after the next congressional hearing or government memo.
Azure Hosting Solves the Data Path, Not the Trust Problem
Microsoft’s strongest argument for a DeepSeek-backed Cowork option is that Azure can contain the operational risk. If the model is hosted by Microsoft, customer prompts and files can remain inside Microsoft’s cloud. Enterprise controls can govern retention, access, auditability, residency, and compliance obligations. That is a very different posture from employees pasting corporate data into a public chatbot.For many commercial customers, that may be enough. A bank, manufacturer, or software company might decide that a cheaper Azure-hosted model is acceptable for low-sensitivity work, especially if Microsoft provides clear controls for model selection, logging, and policy enforcement. A tenant admin could allow less expensive models for summarization and drafting while restricting regulated data to approved models.
But the hard cases will define the controversy. Government agencies, defense contractors, critical infrastructure operators, and heavily regulated enterprises may not care that the workload stays in Azure if the underlying model lineage is politically radioactive. Their question will not be “Where is the prompt processed?” but “What model behavior, training provenance, and legal exposure are we accepting?”
This is where Microsoft’s multi-model strategy becomes both powerful and messy. Customers want choice because no single model is best at every task or price point. They also want someone else to absorb the complexity of deciding which models are safe, compliant, performant, and cost-effective. Microsoft wants to be that abstraction layer, but abstraction does not eliminate accountability.
If Cowork quietly routes work across different models, admins will demand visibility. If Microsoft exposes model choice directly, admins will need policy tools. Either way, model governance becomes part of Microsoft 365 administration. The AI era is turning the tenant admin into a model-risk manager.
Usage Pricing Turns Copilot From License Management Into FinOps
The practical impact for IT departments is immediate: Copilot Cowork will need budgeting, monitoring, chargeback, and guardrails. That is a different operating model from buying a fixed number of Microsoft 365 Copilot seats and watching adoption dashboards. It moves AI spend closer to Azure consumption management than traditional Office licensing.This will be uncomfortable for organizations that treated Microsoft 365 as predictable overhead. A user-based license can be negotiated, allocated, and forecast with relative confidence. A usage-based agent can surprise you, especially if a small group of power users discovers workflows that run constantly or generate repeated high-context requests.
The likely result is a new internal politics of AI access. Executives will want Cowork enabled because the demos look transformative. Finance will want limits because variable bills are hard to explain. Security will want model restrictions. Legal will want assurances about data use. End users will want the system to work without being nagged about credits.
Admins will be caught in the middle. They will need tools that answer basic questions Microsoft has not historically had to expose in Office: Which users are consuming the most AI compute? Which workflows are expensive? Which models are being used? Which data sources are being grounded? Which tasks failed and retried? Which business unit owns the bill?
The winners will be organizations that treat Cowork like a new class of cloud workload rather than a smarter Clippy. The losers will be organizations that enable it broadly, celebrate early adoption, and only later discover that autonomous office work has an autonomous cost structure.
Microsoft’s Multi-Model Pivot Is a Strategic Retreat From One-Model Theology
For years, the Copilot brand was closely tied in public imagination to OpenAI. Microsoft invested heavily, integrated OpenAI models throughout its products, and used that relationship to leap ahead of rivals in enterprise AI distribution. But the economics of agents make dependence on a handful of premium models increasingly risky.A multi-model strategy gives Microsoft leverage. It can route cheaper tasks to cheaper models, reserve expensive systems for complex reasoning, and pressure model suppliers on price. It can also reassure customers that Copilot is not simply a wrapper around one vendor’s roadmap.
This is not a repudiation of OpenAI or Anthropic. It is a recognition that enterprise AI will look more like cloud infrastructure than a single magical assistant. Different workloads will demand different latency, cost, accuracy, privacy, and compliance characteristics. The model layer will become a portfolio.
That portfolio approach is exactly what Microsoft has been building through Azure AI Foundry, Copilot Studio, Microsoft 365 Copilot, and its broader agent platform. The company wants to be the broker, host, policy engine, billing system, and user interface for workplace AI. In that role, Microsoft benefits when models become interchangeable components.
But interchangeability cuts both ways. If customers learn that a cheaper open model is “good enough” for a large share of office work, Microsoft’s premium AI story becomes harder to price. If customers decide model origin matters more than cost, Microsoft’s bargain option may remain confined to less sensitive workloads. The multi-model future is flexible, but it is not simple.
The Real Product Is the Control Plane
The most important Copilot Cowork feature may not be the agent itself. It may be the administrative machinery around it. In an agentic workplace, the question is not merely what an AI can do. It is who allowed it, which data it touched, which model it used, what it cost, and how the organization can prove all of that later.That is why Microsoft’s broader agent governance push matters. As AI agents become embedded in Microsoft 365, admins will need controls that look familiar from identity, compliance, and cloud management. They will need policy scopes, audit logs, approval flows, spend caps, data boundaries, and model allow-lists.
The old Office admin model was built around users and applications. The new model must account for non-human actors performing work on behalf of users across applications. That changes the risk profile. A human making a mistake in Excel is one thing. An AI agent repeatedly applying a mistaken instruction across files, emails, meetings, and connected systems is another.
Cowork’s move to usage-based pricing reinforces that governance problem because cost and control are now linked. An unrestricted agent is not only a security risk; it is a budget risk. A poorly designed workflow is not only inefficient; it is expensive. A hallucinated detour is not only embarrassing; it may consume billable compute.
For WindowsForum readers, this is the part to watch. Microsoft’s success will not depend only on whether Cowork can draft a better deck or schedule a better meeting. It will depend on whether admins can make agentic work governable at enterprise scale.
The DeepSeek Decision Will Test Microsoft’s Enterprise Nerve
If Microsoft ultimately adds a DeepSeek-based option, expect the company to frame it as customer choice. That is the safest language: optional, hosted on Azure, governed by Microsoft controls, suitable for customers who want a lower-cost model. It lets Microsoft avoid saying that AI costs are forcing uncomfortable compromises.But the market will hear the subtext. Microsoft is searching for cheaper inference because agentic AI is expensive to run. It is willing to consider politically sensitive models because cost matters. And it believes Azure’s trust boundary can turn a controversial model into an enterprise-acceptable service.
That is a bold bet. It assumes customers will distinguish between using DeepSeek as a public Chinese-hosted chatbot and using a Microsoft-hosted model derived from DeepSeek technology. Some will. Others will not. In sectors where vendor risk assessments already move slowly, the phrase “Chinese AI model” may stop the conversation before Azure residency controls can rescue it.
Microsoft can reduce the blast radius by being unusually transparent. It should disclose the model identity, hosting architecture, data handling rules, evaluation results, limitations, and admin controls. It should make model selection visible rather than burying it behind a friendly Copilot label. It should let customers disable entire model families if their policies require it.
The worst version of this rollout would be ambiguity: a cheap Cowork mode with vague model language, unclear routing, and marketing-heavy assurances about trust. The best version would be boringly explicit. In enterprise AI, boring is a feature.
The Windows and Microsoft 365 Crowd Should Read the Fine Print
For everyday Windows and Office users, the immediate effect may be limited. Copilot Cowork is an enterprise tool, not a consumer Windows feature. But Microsoft’s direction of travel is clear. AI features that act on your behalf will increasingly be metered, governed, and segmented by model quality.That means users should expect a more tiered Copilot experience over time. Basic chat, summarization, and drafting may remain bundled into existing plans. Longer-running agents, premium reasoning models, and high-volume automation will increasingly sit behind credits, meters, or enterprise controls. The more the AI does, the more the invoice will look like cloud consumption.
For administrators, this is a reminder to slow down before turning on the shiny new switch. Cowork may be genuinely useful, especially for repetitive knowledge work and cross-document tasks. But broad enablement without policy is a recipe for confusion. Usage-based pricing rewards experimentation at first and punishes unmanaged enthusiasm later.
The same caution applies to model choice. A cheaper model may be perfectly acceptable for low-risk workloads. It may be unacceptable for sensitive legal, government, defense, financial, or regulated use. The right answer will vary by organization, which means Microsoft must give admins the knobs to express that policy.
The Meter Is Now Part of the Microsoft 365 Experience
The concrete message from this shift is that Copilot Cowork belongs in the same planning conversation as cloud spend, data governance, and security operations. It is not just another Office feature rolling out on the roadmap.- Organizations should treat Copilot Cowork as a usage-metered cloud workload, not a predictable per-seat productivity add-on.
- Administrators should demand clear reporting on user consumption, workflow cost, model selection, and agent activity before broad deployment.
- Security and compliance teams should decide in advance whether Azure-hosted open or Chinese-origin models are permitted for any class of corporate data.
- Finance teams should expect AI agents to create uneven consumption patterns, especially among power users and departments automating recurring work.
- Microsoft’s model transparency will be central to trust, because customers cannot govern what they cannot see.
- The cheapest model will not always be the safest model, and the safest model will not always be affordable enough for routine agent work.
References
- Primary source: 디지털투데이
Published: 2026-06-17T00:50:08.955948
Microsoft shifts Copilot Cowork to usage-based pricing, weighs adding DeepSeek models
Microsoft is moving its enterprise AI tool Copilot Cowork to a usage-based pricing model, Axios reported on Monday. It is also weighing self-hosting DeepSeek as a cheaper model option, though adding a Chinese AI company’s model could draw criticism, Axios said. Agent tools repeatedly call AI...www.digitaltoday.co.kr - Related coverage: axios.com
Microsoft explores DeepSeek for Copilot Cowork
Microsoft will also shift to usage-based pricing for the enterprise agent.www.axios.com
- Official source: docs.github.com
Models and pricing for GitHub Copilot - GitHub Docs
See per-token pricing for the models available in GitHub Copilot and reference rates for additional usage across plans.
docs.github.com
- Related coverage: livemint.com
‘Goodbye, Copilot’: Microsoft faces backlash as Github Copilot ends flat-rate AI pricing from June 1 | Mint
Microsoft is changing GitHub Copilot's pricing from flat rate to a usage-based system starting from June 1st.
www.livemint.com
- Related coverage: bighatgroup.com
Copilot Weekly: Usage-Based Billing Lands June 1, 2026 | Big Hat Group Inc.
GitHub Copilot moves to usage-based billing on June 1 with AI Credits replacing premium request units. Plus GPT-5.5 GA, BYOK in VS Code, Visual Studio 18.5 Debugger Agent, faster cloud agent, and a 30× capacity plan.www.bighatgroup.com - Related coverage: windowscentral.com
This is Microsoft's new "Copilot Cowork": An experiment with Anthropic's Claude AI models that plans and delegates your work | Windows Central
Microsoft ships Copilot Cowork to its Frontier program.www.windowscentral.com
- Related coverage: tomshardware.com
Github Copilot customers report up to 100-fold price hikes — AI sticker shock bites as Microsoft switches to usage-based pricing | Tom's Hardware
The AI investment chickens have come home to roost.www.tomshardware.com - Related coverage: techgig.com
Microsoft shifts GitHub Copilot to usage-based AI token pricing, TechGig
Microsoft has switched GitHub Copilot to a usage-based AI token pricing model starting June 1, 2026, replacing the previous fixed subscription. Developers will now pay based on AI compute consumption, potentially leading to less predictable monthly bills.techgig.com
- Related coverage: getburnrate.io
BurnRate Blog — AI Coding Cost Insights
Research-backed insights on AI coding tool costs. Industry data on what developers and teams actually spend on Claude, Cursor, Copilot, and Codex.
getburnrate.io
- Related coverage: github.blog
GitHub Copilot is moving to usage-based billing - The GitHub Blog
Starting June 1, your Copilot usage will consume GitHub AI Credits.github.blog
- Related coverage: arstechnica.com
GitHub will start charging Copilot users based on their actual AI usage - Ars Technica
GitHub says it can no longer absorb "escalating inference cost" from it heaviest AI users.arstechnica.com - Related coverage: onehuman.io
GitHub Copilot Moves to Usage-Based Billing... | OneHuman
GitHub Copilot's per-Credit rate is $0.01 USD. Pro plan: $10/1,000 Credits. Typical agentic sessions consume 3,000–4,000 Credits. The math predicts structural overage.onehuman.io
- Related coverage: koskila.net
GitHub Copilot's Freemium Era Is Over - Koskila.net
A week ago I wrote about how the all-you-can-eat AI buffet is seemingly coming to an end, and how practically every major AI vendor has been tightening the...
www.koskila.net
- Related coverage: aitechconnect.in
Microsoft Copilot Wave 3: Claude, Gemini, and Agentic Cowork Arrive
Microsoft 365 Copilot Wave 3 brings multi-model intelligence (Claude, Gemini alongside GPT), Copilot Cowork for agentic task execution, and usage-based billing to enterprise builders.aitechconnect.in - Official source: news.microsoft.com
Copilot Cowork está ahora disponible a nivel general - Source LATAM
news.microsoft.com
- Official source: microsoft.com
Microsoft 365 Roadmap | Microsoft 365
The Microsoft 365 Roadmap lists updates that are currently planned for applicable subscribers. Check here for more information on the status of new features and updates.www.microsoft.com
- Official source: learn.microsoft.com
What's new in Copilot Cowork | Microsoft Learn
Discover the latest features and improvements in Microsoft 365 Copilot Cowork.learn.microsoft.com - Official source: blogs.microsoft.com
Introducing the First Frontier Suite built on Intelligence + Trust - The Official Microsoft Blog
Today Microsoft is announcing: Wave 3 of Microsoft 365 Copilot Expanded model diversity with Claude and next-gen OpenAI models available today General availability of Agent 365 on May 1 for $15 per user General availability of the new Microsoft 365 E7: The Frontier Suite on May 1 for $99 per...blogs.microsoft.com - Official source: techcommunity.microsoft.com
Available today: Anthropic Claude Opus 4.8 in Microsoft 365 Copilot | Microsoft Community Hub
Expanding model choice in Microsoft 365 Copilot: Anthropic's Claude Opus 4.8 is rolling out today in Copilot Cowork.  
techcommunity.microsoft.com
- Related coverage: techradar.com
Microsoft spent billions on Copilot, but only 3.3% of users are actually paying for the AI tools | TechRadar
Microsoft spent $37.5 billion on AI tools, with limited returns so farwww.techradar.com - Related coverage: tomsguide.com
Biggest Microsoft Build 2026 announcements — agentic AI, RTX Spark Dev Box, GitHub Copilot app, new MAI models, and more | Tom's Guide
All the big news from Microsoft's AI-focused eventwww.tomsguide.com - Official source: cdn-dynmedia-1.microsoft.com
- Official source: nist.gov
CAISI Evaluation of DeepSeek AI Models Finds Shortcomings and Risks
The Center for AI Standards and Innovation at NIST evaluated several leading models from DeepSeek, an AI company based in the People’s Republic of China.www.nist.gov - Official source: azure.microsoft.com
