Claude (formerly Cowork) vs ChatGPT Work 2026: Real Cost of AI That Finishes the Work
Short Answer
Both of the biggest AI labs just converged, within about ten weeks of each other, on the same idea: an agent mode that takes an outcome instead of a prompt and hands back a finished document, spreadsheet, deck or web app — and both bundled it into existing subscriptions rather than selling it separately.
- OpenAI launched "ChatGPT Work" on July 9, 2026 — a Codex-powered agent inside ChatGPT that works for hours across connected apps and returns finished sheets, slides, docs and shareable web apps. It has no separate price; it is gated by which ChatGPT plan you're on and metered against that plan's existing usage allowance. Pro, Enterprise and Edu got it first; Plus and Business followed within days; Free and Go do not include it (Free gets only "limited access" through the desktop app, per OpenAI's own pricing page).
- Anthropic launched Claude Cowork on January 12, 2026 as a separate mode (Max-only at launch, then expanded), then, on September 16, 2026 — five days before this research — announced that Cowork and regular chat are merging into one Claude. There is no longer a mode toggle: you type what you need, and Claude decides whether it's a quick answer or a multi-step task. The "Cowork" brand name is being phased out of the interface. This is rolling out to Pro and Max first, with Team and Free "to follow soon" and Enterprise admins getting at least 30 days' notice before it reaches their accounts.
Neither product has a distinct price tag. The real cost question for both is: which subscription tier do you need to get access at all, and how fast does one agentic task consume your plan's agentic usage allowance — on ChatGPT this allowance is distinct from ordinary chat (see §2); on Claude, Anthropic's own documentation describes chat and Claude Code as sharing one combined pool (see §4).
If you are weighing these bundled agent modes against a standalone agent product, our Manus vs ChatGPT Work vs Claude agent cost comparison covers that alternative.
1. What Each Product Actually Is
ChatGPT Work — officially documented (chatgpt.com/pricing feature table, fetched; OpenAI's own launch framing, corroborated across multiple outlets)
Launched July 9, 2026, alongside the GPT-5.6 model family (Sol/Terra/Luna) and a rebuilt desktop app that merges the old standalone Codex app with ChatGPT itself, putting Chat, Work and Codex in one app on every plan, including Free — though having the app does not grant Work access if your plan doesn't include it. The same week, OpenAI began sunsetting its standalone Atlas browser, folding what it learned into ChatGPT and a Chrome extension; Atlas fully closed August 9, 2026.
What it does: takes a goal ("build a competitive analysis of these five vendors"), pulls context from 1,400+ connectable apps, breaks the job into steps, runs independently for hours, and delivers finished spreadsheets, slides, documents or interactive web apps rather than chat text.
Access by plan (officially documented, chatgpt.com/pricing feature comparison, cross-checked against launch reporting):
| Plan | ChatGPT Work access |
|---|---|
| Free | Limited, desktop app only |
| Go ($8/mo) | Limited, desktop app only |
| Plus ($20/mo) | Full access, "Desktop, web, and mobile" |
| Pro ($100/$200/mo) | Full access, expanded usage |
| Business (formerly Team) | Full access, admin-governed |
| Enterprise | Full access, admin spend controls |
Claude (Cowork merge) — officially documented (claude.com/blog/cowork-is-now-claude, fetched via search snippet; support.claude.com Cowork help article; claude.com/pricing)
Cowork itself launched January 12, 2026 as a Max-only research preview (macOS first), described by Anthropic as "Claude Code for the rest of your work" — folder-permission based, letting Claude read/write/create files in a directory you grant access to, planning and executing multi-step tasks with visible progress tracking. It later added a cloud execution mode (runs on Anthropic's servers in an isolated environment; work continues even after you close your laptop, resumable from any surface) and expanded beyond Max-only.
On September 16, 2026, Anthropic merged Cowork into regular Claude. Per Anthropic's own announcement: "You don't have to choose where a task goes... Bring a quick question, or hand over a report due at noon, and Claude takes it from there, even after you've closed your laptop." The same announcement launched Claude Docs and Claude Slides (beta, paid plans only) and folded Claude Design (previously a standalone tool, built on Canva's engine, launched ~April 2026) directly into ordinary conversations. Claude Code remains explicitly separate — the consolidation does not touch the developer-focused coding tool.
Rollout sequencing (officially documented): Pro and Max plans first, over "the coming weeks," on web/desktop/mobile. Team and Free "will follow soon." Enterprise admins get at least 30 days' notice before the change reaches their org. Existing Cowork users keep their chats, projects, artifacts, connectors and Skills — nothing is lost, per Anthropic.
Access by plan (officially documented, claude.com/pricing):
| Plan | Claude Code (unaffected by merge) | New unified "Claude" (Cowork capability) |
|---|---|---|
| Free | No | Not yet (Cowork capability rolling out to Pro/Max first) |
| Pro ($17–20/mo) | Yes | Yes — first wave |
| Max (from $100/mo) | Yes | Yes — first wave |
| Team (Standard $20/25, Premium $100/125 per seat) | Yes | Rolling out "soon" (not yet at fetch time) |
| Enterprise | Yes | 30+ days' notice before rollout |
2. Pricing Mechanics: Neither Product Has Its Own Meter
This is the central finding of this article, and it holds for both companies, though the pool structure differs between them: neither ChatGPT Work nor the unified Claude agentic mode is sold as a discrete product with its own price — but "shares a pool" means something different on each platform. On ChatGPT, ordinary text chat is metered separately from the agentic Work/Codex allowance; on Claude, chat and Claude Code are explicitly documented as drawing from one combined pool.
- OpenAI, explicit (officially documented, help.openai.com — "Using Codex with your ChatGPT plan" and "ChatGPT Work and Codex"): "Codex, ChatGPT Work, ChatGPT for Excel, and Workspace Agents use a shared allowance and credit pool when those features are available on your plan." Crucially, OpenAI's own Business rate-card documentation states this allowance is separate from ordinary Chat usage: "Chat usage is metered separately from Astra usage in Work and Codex," and a support article on hitting Work's limit confirms "Regular Chat keeps working while Work is paused." ChatGPT Work and Codex share one agentic usage allowance where supported; ordinary text chat draws from a separate meter and should not be described as consuming that same Work/Codex pool. There is no separate dollar price for this allowance — tasks consume a variable slice of it depending on complexity, and OpenAI has not published a per-task rate.
- Anthropic, explicit (officially documented, claude.com/pricing FAQ): "Claude Code is included in all paid plans. It shares the same usage limits as the rest of your plan, so your work in the terminal and your chats draw from one pool." Anthropic's own usage-limits FAQ states that "activity across Claude on web, desktop, mobile, and Claude Code all draws from the same pool." This documented pool-sharing covers chat and Claude Code specifically; the September 16, 2026 merge announcement does not say whether the newly merged agentic task capability (formerly Cowork) draws from that pool or from a separate allowance. The documented sharing therefore applies to chat and Claude Code; the task mode's pool relationship remains an open question. No fixed message or task count is published for any Claude tier; limits are described only as multiples of each other (Pro ≥5× Free per 5-hour session; Max 5x/20x more than Pro).
Practical consequence: you cannot compute "$X per finished report" the way you can for a metered API. The real cost is whichever subscription tier the feature requires, plus the opportunity cost of a long agentic task consuming usage headroom that would otherwise go to ordinary chat, coding, or (on Claude) Design/Docs/Slides work.
3. Subscription Cost to Get Access (CALCULATION, from confirmed plan prices)
| Route | Cheapest plan that grants full access | Monthly cost |
|---|---|---|
| ChatGPT Work, individual | Plus | $20 |
| ChatGPT Work, team | Business Standard seat | $20/seat (annual) / $25 (monthly) |
| Claude unified agent mode, individual | Pro | $17 (annual) / $20 (monthly) |
| Claude unified agent mode, team | Team Standard seat, once rollout reaches Team | $20/seat (annual) / $25 (monthly) — not yet live for Team at fetch time |
At the individual level, the two products cost the same $20/month headline (ChatGPT Plus vs Claude Pro monthly), with Claude's annual price ($17/mo) undercutting Plus's flat $20 (ChatGPT Plus has no published annual discount, per its own pricing page structure). At the team level, the comparison is currently incomplete: ChatGPT Work is live on Business today; Claude's equivalent capability has not yet reached Team plans as of September 26, 2026, so a same-tier team cost comparison cannot yet be made — this is a timing gap, not a pricing gap.
Cheapest way to try either without a real subscription: OpenAI's Free/Go tiers get only "limited" Work access via the desktop app; Anthropic's Free tier does not yet have the unified agent mode at all (it is Pro/Max-first). Neither company currently offers a genuinely free way to use this specific capability at meaningful volume.
4. What Draws From Which Pool — and Why That Matters for Cost
- On ChatGPT: Work shares its agentic allowance with Codex specifically (and, per OpenAI's Business documentation, with ChatGPT for Excel/PowerPoint and Workspace Agents where available) — not with ordinary text chat, which OpenAI explicitly meters separately. A single long agentic Work task — "gather context across five apps and produce a slide deck" — can consume a materially larger slice of that Work/Codex allowance than a single Codex coding task, using the same "variable cost by complexity" logic OpenAI applies throughout. Practically: you can exhaust your Work/Codex allowance for the month and still use ordinary ChatGPT chat normally, confirmed by OpenAI's own support content ("Regular Chat keeps working while Work is paused").
- On Claude: chat and Claude Code are documented as sharing one pool. Anthropic's September 16, 2026 merge announcement does not say whether the newly merged agentic task mode draws from that same combined pool or a separate allowance. If it shares the documented chat/Code pool (an inference, not a stated rule), a multi-hour report-and-slides task alongside Claude Code work would draw from one combined budget.
- Neither company publishes a way to estimate in advance how much of your allowance a given task will consume — OpenAI's Work page and Anthropic's usage FAQ both describe cost only qualitatively ("variable," "depends on complexity and length"), which is the single biggest practical uncertainty for anyone trying to budget around either product.
5. Feature Comparison (Not Cost, but Cost-Adjacent)
| Capability | ChatGPT Work | Claude (unified) |
|---|---|---|
| Takes an outcome, works for hours, returns finished output | Yes | Yes |
| Native document/slide creation | Via connected apps and file generation | Claude Docs and Claude Slides (new, beta, launched with the merge) — direct in-conversation creation, edit, present, export to PowerPoint/PDF |
| Design tool integration | Not specifically named in research | Claude Design (Canva-engine based) now works inside conversations |
| Continues after you close your laptop | Not stated in the cited OpenAI documentation | Confirmed — cloud execution mode, resumable from any surface |
| Connects to external apps/tools | 1,400+ connectable apps via Plugins directory | Connectors, Skills (carried over from Cowork) |
| Scheduling | Not confirmed in the cited public information | Confirmed — "Schedule it for every Monday, and Claude starts on the report without being asked" |
| Coding-specific agent, kept separate from this mode | Codex (shares the same app, distinct mode) | Claude Code (explicitly not merged) |
Both vendors emphasize checkpoints/approval: OpenAI frames Work's oversight as standard chat-style review of deliverables; Anthropic's Claude lets you choose whether it asks before every action or only checks in "when something needs a closer look" — a configurable autonomy level.
6. Hidden Costs
- Neither product publishes a per-task or per-hour cost, which makes budgeting inherently approximate. Unlike the API-priced comparisons elsewhere on NotCheapAI, there is no rate card to compute against — the only lever is which subscription tier you buy and how conservatively you use the agent mode.
- Timing risk for teams: if your organization is on Claude Team specifically to get this capability today, it is not yet available there — you would need individual Pro or Max seats in the interim, which may cost more per person than a Team seat once the feature does arrive.
- Enterprise rollout lag is explicit and lengthy on Claude (30+ days' notice) — organizations planning around this feature on an Enterprise contract should not assume immediate access.
- "Limited access" on ChatGPT Free/Go is undefined in concrete terms — OpenAI has not published what "limited" means in task count or complexity, so Free/Go users cannot reliably budget time around it.
- Atlas's shutdown (August 9, 2026) removes a separate browsing product some workflows may have depended on; OpenAI states the capability folded into ChatGPT itself and a Chrome extension, but any workflow built specifically around the standalone Atlas app needs to be re-verified against the current ChatGPT Work / Chrome extension combination.
- Claude Design, Docs and Slides are explicitly beta — feature completeness and reliability should be expected to shift during the beta period, which is not itself a "cost" but affects how much production work should be routed through them yet.
7. BUILD & EARN Scenario: A Report-Production Service
MODELING ASSUMPTION: a consultancy delivers 20 finished client reports/month (research + written summary + slide deck), billed at $150/report = $3,000 revenue, using either ChatGPT Plus or Claude Pro as the production tool for one operator. Metric: platform-cost share of revenue (excludes the operator's own labor reviewing and finalizing each report — not profit).
| Route | Subscription cost | Share of $3,000 |
|---|---|---|
| ChatGPT Plus (1 seat) | $20/mo | 0.67% |
| Claude Pro (1 seat, annual) | $17/mo | 0.57% |
At this volume, the subscription cost is trivially small on either platform — the entire economic question is whether one $20/month seat's agentic usage allowance (ChatGPT's separately-metered Work/Codex pool, or Claude's combined chat/Code/task pool) can actually sustain 20 multi-hour agentic report-generation tasks a month without hitting undisclosed usage limits, not which platform is "cheaper." Since neither vendor publishes a task-count ceiling, the only way to validate this in practice is to run the actual workload and watch for rate-limit or "upgrade to continue" prompts — this article cannot answer that question from published pricing alone.
FAQ
Does ChatGPT Work cost extra on top of my ChatGPT plan? No — it has no separate price. Access is gated by plan (Plus and above for full access), and usage draws from your existing plan allowance.
Is Claude Cowork still a separate product? As of September 16, 2026, no — Anthropic merged it into regular Claude. There's no more toggle between "Chat" and "Cowork" on accounts that have received the update; Claude decides internally whether a request is a quick answer or a longer task.
Which is cheaper, ChatGPT Plus or Claude Pro, for this kind of work? Nominally the same at $20/month billed monthly; Claude Pro is cheaper on annual billing ($17/month) since ChatGPT Plus has no published annual discount.
Can my team use this today? ChatGPT Work is live on Business (OpenAI's team plan) today. Claude's equivalent capability is Pro/Max-only as of September 26, 2026 and has not yet reached Team plans.
Is Claude Code affected by the Cowork merge? No — Anthropic explicitly kept Claude Code as a separate, unmerged product for developers.