Claude Cowork vs Copilot Cowork vs ChatGPT Work: practical tasks, prices and how to choose
Compare real business tasks across three AI working tools: meeting briefs, reports, supplier research and proposals. Understand included usage, extra charges and how to run a useful trial.
By Ivan Pylypchuk, CEO of Softblues. Sources checked 28 September 2026.
Claude Cowork, Microsoft Copilot Cowork and ChatGPT Work can take a task beyond a chat answer and produce working files. For a UK company, the useful comparison starts with the job: where its information lives, what must be produced and who checks it. Price matters, but the subscription alone does not tell you the cost of completing that job.
Softblues is a London software company helping UK and Irish firms become AI-native. We work across these platforms. This research compares documented capabilities and buying terms; the scenarios below are suggested tests, not a controlled performance benchmark.
Key facts
What is Cowork, and why would a company use it?
Think of a monthly report. A chat assistant can suggest its structure or rewrite a paragraph. A task-working assistant can be asked to read the approved inputs, build a spreadsheet, draft the report and return files for review. The value is in reducing the manual movement between sources and deliverables.
“Cowork” is not one shared product or an industry standard. Anthropic and Microsoft use it in product names; OpenAI calls its comparable task-working experience ChatGPT Work. Their capabilities overlap, but the available connections, execution environments and billing rules differ.
These tools are useful when the work has identifiable inputs and a reviewable output. Preparing a meeting brief, turning notes into a proposal or assembling a management pack are sensible starting points. A task involving an unclear business decision still needs a person to define what a good result looks like.
They also need instructions that describe the work. “Analyse these files” leaves much unresolved. “Use these four approved files to prepare a two-page report; show missing information; do not invent figures; return the workbook behind every total” gives the reviewer something concrete to accept or reject.
Which tasks should you compare across the three platforms?
The table gives practical routes to test, not exclusive capabilities or quality rankings. Connections must be enabled and authorised. All three may solve the same job; the difference can be how much preparation and checking your team needs.
| Business task and useful output | Copilot Cowork: route to test | Claude Cowork: route to test | ChatGPT Work: route to test |
|---|---|---|---|
| Prepare a client meeting: a brief, open questions and draft follow-up | Use relevant Microsoft 365 emails, meetings and shared files as context. Ask for a brief with links back to evidence. | Provide approved notes and documents, or use enabled connectors. Apply a reusable briefing template. | Supply source files and authorised connected context. Request a sourced brief and draft follow-up. |
| Produce a monthly management pack: workbook, narrative and slides | Work from shared Office files; test updates to existing Word, Excel and PowerPoint files. | Give access to the source folder, reporting instructions and template; request working files with calculations. | Combine approved files with the reporting brief; request an analysis workbook and presentation. |
| Compare suppliers: evidence matrix and recommendation | Use proposals and internal requirements in Microsoft 365; request a comparison document for review. | Read a folder of proposals against a fixed checklist; surface omissions and conflicting terms. | Combine proposals with permitted research; produce a comparison workbook with sources and unanswered questions. |
| Draft a proposal: document ready for commercial review | Combine meeting context and the approved Office template; test editing the shared document. | Apply house instructions to the discovery notes and template; flag missing scope and assumptions. | Turn the brief and supplied evidence into a document or deck; separate evidence from proposed recommendations. |
These routes are grounded in the vendors' task and file capabilities. They are our test designs, not claims that a particular platform has completed these scenarios with measured results. Microsoft tasks Claude tasks OpenAI tasks
For a management pack, the review should check that totals reconcile and every narrative claim has a source. For a supplier comparison, check that missing evidence is marked “unknown”, rather than converted into an invented answer. Those checks often tell you more than how polished the first page looks.
Where does each platform make a sensible starting point?
Copilot Cowork: start with work already inside Microsoft 365. Microsoft documents context from email, meetings, Teams and files, alongside document creation and actions. It can edit existing Word, Excel and PowerPoint files in OneDrive or SharePoint when the user has edit permission, preserving the shared file and version history. That is useful when the final deliverable belongs in an established Office workflow. Use Cowork
Check which actions need approval in your configuration. Do not assume every email, file change or external action will always wait for a separate confirmation. Ask IT to demonstrate the permitted scope and the consumption report before inviting the pilot group.
Claude Cowork: start with a folder and a repeatable procedure. Claude documents direct local file access on desktop, working spreadsheets, presentations, formatted documents and reusable instructions. A reporting folder plus a house template is a useful trial. Cloud sessions and connectors offer other routes; do not assume their access is identical to desktop access. Get started
Test whether it follows your house procedure over several runs, including one with missing input. Team and Enterprise administration differ, and their current business-plan experience should not be inferred from announcements about personal Pro or Max plans. Business availability
ChatGPT Work: start with research that must become a deliverable. OpenAI describes multi-step tasks using files and connected tools, including comparison workbooks and presentations. A supplier assessment is a useful trial: give it the criteria, permitted sources and required output, then inspect whether its recommendation follows the evidence. Get started with Work
Work can run in cloud or local environments with different permissions. A local task does not establish that all messages and context remain on the device. Check the actual route for the material you plan to use. Work and Codex
These are starting points, not boundaries. Claude can research, ChatGPT can work with document packs, and Microsoft can produce comparisons. If two platforms fit, run the same task in both before declaring a winner.
Is Copilot Cowork included in the Microsoft subscription?
No included Cowork allowance comes with the Copilot subscription. Microsoft's licensing guidance explicitly distinguishes the prerequisite licence from Cowork's usage charge. Buying a licence therefore does not give a team a bundled quantity of Cowork tasks. Copilot Credits guidance
Cowork consumes Copilot Credits according to the work performed. Complexity and the combination of models, runtime, context and tools affect consumption. Credits can be purchased through supported prepaid or pay-as-you-go arrangements. A separately purchased pool is not an allowance included in the seat.
This does not mean every Microsoft AI feature is charged in the same way. Everyday Copilot capabilities and included agents have their own entitlements. Do not apply Cowork's billing rule to the entire Microsoft product family.
For a pilot, record the actual credit consumption of representative tasks. A simple brief and a long task involving many files should not be assumed to cost the same. Set the spending policy before the trial, then budget from observed use rather than an invented fixed price per report.
How do subscription prices compare?
Published prices checked on 28 September 2026 are below. Annual prices are monthly equivalents with an annual commitment. Dollar prices remain in dollars; obtain a dated sterling quote and applicable tax treatment rather than treating a currency conversion as the supplier's UK price.
| Plan | Published seat price | What happens to agent usage? |
|---|---|---|
| Microsoft Business Premium with Copilot | £24.60/user/month, paid yearly, excluding VAT | Full Microsoft 365 bundle, not AI-only pricing. Cowork usage is additional, with no included Cowork entitlement. Microsoft UK pricing, checked 28 September 2026 Microsoft Copilot Credits licensing guidance |
| Claude Team Standard | $20/user/month annually; $25 monthly | Includes a limited usage allowance; multi-step work consumes it. Additional usage is separate where enabled. Claude pricing Get started with Claude Cowork |
| Claude Enterprise | $20/user/month annually, plus usage | Seat charge plus usage at API rates; do not assume the Team allowance carries over. Claude pricing |
| ChatGPT Business Standard | $20/user/month annually; $25 monthly | Work is included with a usage allowance shared with Codex; eligible extra usage uses credits. ChatGPT Work and Codex ChatGPT Business FAQ |
| ChatGPT Enterprise | Contract quote | Confirm included usage, controls and overage in the agreement. ChatGPT Enterprise pricing |
For 100 seats, those annual Standard rates give Claude Team or ChatGPT Business a $2,000 monthly seat equivalent. The Microsoft bundle gives £2,460 monthly, but includes wider Microsoft 365 services. For an existing Microsoft customer, compare the incremental licence cost against the current estate, then add Cowork usage. The full bundle is not a like-for-like AI surcharge.
A useful budget is incremental licences + measured usage + setup and support + review time. Our business-case method explains how to connect released capacity to an operational result without calling every saved minute cash.
What should a UK firm check before using confidential files?
Permissions, storage and processing are separate checks. Microsoft documents UK storage commitments for eligible UK-default tenants, while files follow their service terms. That does not mean every model processes every request in the UK. Microsoft residency
Anthropic describes US storage for first-party data. OpenAI offers UK storage for eligible Enterprise content, but does not list UK inference residency. Connected services require their own checks. Anthropic OpenAI
Record the permitted data, retention, access and approval route with the responsible IT and privacy owners. Residency is a documentation task, not a legal barrier by default; a client's particular contractual restrictions still matter.
How should you choose without buying three platforms for everyone?
Start with one recurring task and one accountable reviewer. Use a non-sensitive example with realistic complexity. Define the required output before opening the product.
1. Prepare identical inputs. Include the template, source files and a deliberately missing item. Record any preparation each platform needs. 2. Run the task in the existing environment first. For a Microsoft-first firm, test Copilot where the workflow fits, after enabling and capping usage billing. 3. Compare a second platform where there is a reason. Test Claude for your document procedure or ChatGPT Work for your research-to-deliverable workflow. These are hypotheses to test, not predetermined winners. 4. Measure accepted output. Record missing evidence, formula errors, reviewer corrections, manual steps and actual consumption. Include unsuccessful runs. 5. Choose the smallest workable deployment. Keep a governed assistant seat if it solves the job. Add integration or an engineered agent when the process requires it.
Our platform framework distinguishes adoption-first support for people from automation-first delivery of a workflow. Softblues' own Claude Operating System case study shows our operating approach, not proof that every firm should make the same choice.
Get the Copilot and Claude deck, or book a discovery call to define your first practical comparison.
Frequently asked questions
Is Cowork just another chatbot?
It can work through a task and return files, rather than only an answer. The useful distinction is whether it completes your defined steps with the authorised inputs and tools.
Is Copilot Cowork usage included in a Copilot licence?
No. Microsoft requires the licence and charges Cowork usage separately, without a bundled Cowork entitlement. Other Copilot features have different rules. Microsoft Copilot Credits licensing guidance
Do Claude and ChatGPT include unlimited agent work?
No. Claude Team and ChatGPT Business include allowances with limits. Claude Enterprise adds metered usage to its seat charge; ChatGPT Enterprise terms depend on the contract. Claude pricing ChatGPT Business FAQ
Which task should a company try first?
Choose a recurring task with clear inputs and an output someone can check: a meeting brief, report, supplier comparison or proposal. Avoid starting with an autonomous process that makes commitments without a reviewer.
Dated sources
Vendor documentation checked on 28 September 2026. Use-case recommendations are Softblues' analysis; no comparative performance benchmark is claimed.
See it in production
Systems we have built and run for clients, with the numbers that came out of them.
Related Articles
Claude Enterprise vs Team vs Pro: Which Plan Does Your Business Need?
Gemini for Business vs Claude Enterprise: Which Fits Your Company in 2026?
