A decision-oriented comparison of Google Gemini and Gemini Notebook, ChatGPT, Claude, and Microsoft 365 Copilot across their lowest paid personal tiers and practical power-user tiers.
Entry plans optimize access. Roughly $20 plans optimize execution. The right purchase depends on whether the user needs a conversational assistant, an integrated productivity suite, or an agentic workbench.
The highest-value pairing is usually complementary rather than same-category. Combine one reasoning/workflow system with one ecosystem or media system.
The report is rebuilt around changes that affect price-to-capability, surface availability, naming, memory behavior, or workflow portability.
Fable 5 uses usage credits from the first request on Pro and Team Standard. Max includes it within a constrained share of weekly usage.
The product remains standalone. Secure cloud-computer code execution is available now to Google AI Ultra and eligible Workspace customers, with Pro rollout announced as upcoming.
Plus and higher plans can store up to 5,000 characters. Go is not listed for this increase.
Users can search chats, Projects, images, and documents across web, iOS, and Android.
Individual categorized entries replace the daily memory summary for the new consumer memory experience.
Work handles long-running deliverables and scheduled tasks. The Plugin Directory replaces the App Directory as the workflow-discovery layer. Web and mobile Work exclude Free and Go.
Remote sessions can continue without a device online, beginning with Max and expanding later. Local Dispatch work still requires the desktop to remain awake.
“Agent Mode” is being integrated as Edit with Copilot. Deep Research is retired; Researcher is the Premium in-depth research agent. Word, Excel, and PowerPoint Agents remain available to Personal.
US public list prices before tax and promotions. Annual equivalents are shown only where annual billing exists.
400 GB. Two-times Gemini usage versus non-AI plans. Pro-model access, Deep Research, media generation, Gemini Notebook, Daily Brief, Omni, and AI Inbox rollout.
More messages, uploads, images, analysis, memory, and context. The Go help page documents Projects, Tasks, custom GPTs, and Library subject to availability.
$99.99/year or $9.99/month. Desktop Office apps, 1 TB OneDrive, 25 agent tasks monthly, 60 image credits, and Word/Excel/PowerPoint Agents.
$200/year or $20/month. Code, Cowork, Design, Science, Research, unlimited Projects, Skills, connectors, and Microsoft 365 add-ins.
| Platform | Practical tier | Price | What changes | Important caveat |
|---|---|---|---|---|
| Google AI Pro | $19.99/month | Four-times usage, 5 TB, expanded Gemini Notebook and media, AI Studio, Antigravity, Jules, YouTube Premium Lite. | Features vary by country, age, language, and surface. | |
| OpenAI | ChatGPT Plus | $20/month | GPT-5.6 reasoning, expanded deep research, images, Codex, Work, Projects, Tasks, GPTs, Plugins. | Go documentation lists some Plus-table features; Work explicitly excludes Go on web/mobile. |
| Anthropic | Claude Pro | $17 annual equivalent or $20 monthly | Code, Cowork, Design, Science, Research, Skills, connectors, Projects. | Newest Fable 5 usage is metered separately on Pro. |
| Microsoft | Microsoft 365 Premium | $16.67 annual equivalent or $19.99 monthly | Researcher, Analyst, extensive AI usage, advanced AI features, up to 6 TB for six people. | AI features remain limited to the subscription owner. |
Status reflects the entry paid plan, not the vendor’s premium or business showcase.
| Capability | Google AI Plus | ChatGPT Go | Claude Pro | Microsoft 365 Personal |
|---|---|---|---|---|
| Persistent workspace | Yes Gemini Notebook and notebooks in Gemini Apps | Yes Projects available to all plans | Yes Unlimited Projects | Yes Notebooks and Pages |
| Cross-chat memory | Yes Memory from chats and connected Google services | Yes Memory summary, sources, files, and connected apps | Yes Categorized entries and paid past-chat search | Yes Saved memory and chat-history inference |
| Reusable assistant | Yes Gems | Documented Custom GPTs subject to availability | Yes Skills, Projects, plugins | Yes User-built and Office agents |
| Portable behavior package | Vendor-specific | Partial Plugins may bundle skills, apps, templates | Strongest Agent Skills and MCP | Vendor-specific |
| Scheduled tasks | Yes Scheduled actions | Documented Tasks; Work schedules require Plus | Yes Cowork schedules; local versus cloud matters | Limited 25 agent tasks; general scheduling remains fragmented |
| Deep research | Yes More access | Limited Plus expands it | Yes Research included | No Researcher requires Premium |
| Terminal coding agent | Adjacent Jules and Antigravity vary by plan | Limited Expanded Codex on Plus | Yes Claude Code included | Not central |
| Native raster image generation | Strong Image, music, video | Strong More than Free | Structured SVG, diagrams, design—not a conventional photo generator | Credits 60/month |
| Editable UI/design workflow | Canvas | Canvas Work requires Plus | Best Design + Code + artifacts | Office-native Best for finished Office artifacts |
| Business governance | Separate Workspace product | Business or Enterprise | Team or Enterprise | Business licensing |
Each dossier separates the cheapest plan from the tier that materially changes professional or technical workflows.
Best bundle economics and strongest low-cost media/research ecosystem.
Broadest general-purpose assistant and the most complete all-around upgrade path.
Best personal workflow composition and design-to-code continuity.
Best when the end state is a real Office file, not merely a chat answer.
A useful comparison must distinguish explicit user profile, inferred personalization, past-chat retrieval, project context, source retrieval, and agent execution state.
Stable facts and preferences applied across conversations. Best for personalization; risky as an authoritative record.
Search or RAG over prior conversations. Claude exposes it as a visible tool call; ChatGPT now exposes unified search.
Projects, Notebooks, files, and instructions. The most dependable layer for repeatable work.
Email, documents, calendar, photos, Slack, and other services. High value, but deletion and authorization boundaries become more complex.
Schedules, task history, approvals, intermediate files, and outputs. Often not portable to ordinary chat memory.
Versioned files, sources, and decisions outside the assistant. This should remain the system of record.
| Criterion | Gemini | ChatGPT | Claude | Microsoft 365 Copilot |
|---|---|---|---|---|
| User-visible memory | Settings and activity controls | Editable memory summary and legacy saved memories | Categorized entries | Saved memories and personalization settings |
| Past-chat search | Memory can reference activity; interface differs by surface | Unified manual search; memory may synthesize past chats | Explicit paid search tool across non-project chats or within one project | Chat-history inference; application context varies |
| Source explanation | Personalization settings; not one universal ledger | Book icon can show memory, chats, files, instructions, and apps | Past-chat searches appear as tool calls | Less unified across standalone and embedded surfaces |
| Portability | Can import memory/chats; data follows Gemini activity rules | No general vendor-neutral memory export workflow documented | Memory import and export | No equivalent portable memory package |
| Deletion complexity | Delete chats and disconnect relevant apps | Delete every contributing chat, file, memory item, and app source | Edit/delete entries; project and incognito scope remain separate | Controls distributed across Copilot and Microsoft 365 surfaces |
Assess where execution occurs, which tools are available, whether the device must remain awake, what approval gates exist, and how failures are surfaced.
| Platform | Entry scheduling | Power workflow | Runs without device? | Action boundary |
|---|---|---|---|---|
| Google AI Plus | Scheduled actions | Higher tiers add Spark and more agentic Search/Antigravity capabilities | Yes for supported cloud actions | Google services, account permissions, and regional feature availability |
| ChatGPT Go | Tasks documented subject to availability | ChatGPT Work on Plus and higher adds schedules, triggers, monitoring, apps, and deliverables | Yes for supported cloud Work/tasks | Work approvals, apps, files, and plan-specific tools |
| Claude Pro | Cowork scheduled tasks | Dispatch, computer use, Code, connectors, plugins, local and cloud sessions | Cloud session: yes. Local Dispatch: no. | Highest personal action reach; requires strongest safety controls |
| Microsoft 365 Personal | 25 agent tasks shared monthly | Premium Researcher, Analyst, extensive usage, and advanced agents | Depends on the agent and Office service | Native Office operations; organizational automation is a separate tier |
A feature can exist on both plans while identity, retention, audit, connector approval, policy, and training defaults differ materially.
Personal strength: remote MCP, desktop extensions, Skills, plugins, Code, Design, and Cowork.
Business unlock: SSO, connector controls, organization-wide Skills, audit, compliance API, retention, and roles.
Personal strength: Apps, Plugins, Projects, GPTs, deep research, Codex, and Work on eligible plans.
Business unlock: company knowledge, managed workspace agents, custom MCP governance, policy, administration, and deployment.
Personal strength: deep Google ecosystem context and family sharing on Pro.
Business unlock: Workspace is a separate identity and entitlement architecture. Consumer Google AI plans cannot be purchased by Workspace accounts.
Personal strength: real Office files and native application operations.
Business unlock: Microsoft Graph, Teams, SharePoint, Copilot Studio, managed agents, compliance, and organizational controls.
Raster generation, editable visual design, implementation code, and finished application-native artifacts are different production jobs.
Best low-cost breadth for image, music, video, Flow, multimodal reference work, and research-grounded creative exploration.
Best general iterative loop across research, images, writing, code, Canvas, Work, and Sites when using Plus.
Best structured design and UI handoff through Design, artifacts, SVG, HTML, Skills, MCP, and Claude Code.
Best finishing environment for PowerPoint, Word, Excel, Designer, Brand kits, and native document collaboration.
| Output | Primary | Why | Secondary | Cost note |
|---|---|---|---|---|
| Illustration or visual concept | Google AI Plus | Lowest subscription with broad image and media tools | ChatGPT Go/Plus | Google leads cost-adjusted breadth |
| Editable UI concept | Claude Design | Design-system extraction and implementation continuity | ChatGPT Canvas/Work | Claude costs more but reduces handoff loss |
| Production HTML or React | Claude Code | Skills, MCP, repository, and design context | ChatGPT Codex | Both may consume shared quotas quickly |
| Research-grounded learning artifact | Gemini Notebook | Explicit source corpus and notebook outputs | ChatGPT deep research | Code execution is not yet a Plus entitlement |
| Finished presentation | Microsoft 365 Copilot | Native PowerPoint, templates, Brand kits, and iterative editing | Claude Design export | Personal is enough for PowerPoint Agent; Researcher needs Premium |
| Structured SVG and diagrams | Claude | Code-native, editable output | ChatGPT | No raster-generation requirement |
Screenshots, breakpoints, component inventory, tokens, interactions, and states.
Use Claude Design or ChatGPT to infer structure, tokens, and component rules.
Use Google or ChatGPT for illustrations, references, and visual variants.
Build semantic components with Claude Code, Codex, or Gemini coding tools.
Automate screenshots, overflow, contrast, keyboard, touch, and localization tests.
The latest releases improve continuity, but the most capable surface still differs by platform.
| Platform | Web | Desktop unlock | Mobile unlock | Parity risk |
|---|---|---|---|---|
| Gemini | Gem creation, Canvas, Gemini Notebook, AI Studio, repositories | Browser-centric; Chrome auto-browse and developer tools vary by tier | Android actions, camera, Live, calls, messages, Home, Photos, location | Features vary heavily by account, region, device, and app |
| ChatGPT | Projects, GPTs, search, Work, Plugins, Canvas, research | New Chat/Work/Codex structure, local versus cloud Work, Projects, richer coding | Voice, camera, images, unified search, Work continuation, Codex Remote approval | Work excludes Go on web/mobile; Codex selection remains desktop-centric |
| Claude | Chat, Projects, Design, remote Cowork rollout, artifacts | Code, Cowork, computer use, local files, extensions, Office and browser integration | Dispatch, remote task assignment, persistent Cowork thread, interactive connectors | Remote Cowork rollout starts with Max; local execution needs desktop awake |
| Microsoft | Copilot app, Pages, Notebooks, Office web, agents | Full Word, Excel, PowerPoint, Outlook, Brand kits, formatting and offline work | Conversational Copilot and lighter Office capture/review | Feature sets vary by app and OS; Excel has OneDrive/AutoSave requirements |
These combinations minimize overlap while increasing validation, media breadth, workflow execution, or native artifact quality.
Google AI Plus: Gemini, Gemini Notebook, Google context, media generation, and 400 GB.
ChatGPT Go: higher usage, files, images, memory, Projects, and documented access to Tasks and GPTs.
Google AI Plus + ChatGPT Go. Strong research, media, memory, Projects, and cross-validation.
Claude Pro annual + Google AI Plus. Skills, MCP, Code, Design, Cowork, Notebook, image, music, and video.
Claude Pro annual + Microsoft 365 Personal annual. Build and reason in Claude; finish as native Office files.
All four entry paid plans. Appropriate for portability standards, procurement research, and repeated cross-platform evaluations.
Every vendor-specific GPT, Gem, Skill, Notebook, Plugin, or agent should be generated from a canonical package rather than becoming the only copy of the workflow.
ai-capability-pack/
├── manifest.yaml
├── instructions/
│ ├── system.md
│ └── workflow.md
├── knowledge/
│ ├── sources/
│ ├── claims.jsonl
│ └── glossary.md
├── skills/
│ ├── SKILL.md
│ ├── scripts/
│ └── examples/
├── design/
│ ├── tokens.json
│ ├── components.md
│ ├── states.md
│ └── reference-screens/
├── tools/
│ ├── mcp/
│ └── openapi/
├── evals/
│ ├── cases.jsonl
│ ├── expected/
│ └── red-team/
└── adapters/
├── chatgpt/
├── claude/
├── gemini/
└── microsoft/
The rebuild treats the product pages as changing evidence, not as stable contracts.
| Evaluation dimension | Weight | Measurement |
|---|---|---|
| Task completion accuracy | 20% | Human-reviewed accepted result against requirements |
| Source and citation fidelity | 15% | Correct attribution, coverage, and no unsupported claims |
| Artifact editability | 10% | Native structure retained after generation |
| Workflow repeatability | 10% | Same input produces stable process and recoverable state |
| Context retrieval | 10% | Precision and recall across controlled canaries |
| Cross-surface consistency | 10% | Web, desktop, mobile, and embedded applications |
| Total cost per accepted artifact | 10% | Subscription allocation plus usage credits and correction time |
| Portability | 5% | Neutral export, Skills, MCP, source and design reuse |
| Action safety | 5% | Authorization, confirmation, provenance, and rollback |
| Mobile usability | 5% | Capture, control, continuation, and small-screen recovery |
Confirmed product direction is separated from informed forecast.
Accessed July 21, 2026. Product pages, account entitlements, prices, models, limits, and rollouts may change without notice.