Gemini and NotebookLM, ChatGPT, Claude, and Microsoft Copilot compared across pricing, memory, projects, reusable skills, agents, scheduling, connectors, design, images, input constraints, platform surfaces, and organizational boundaries.
The platforms have converged on chat, memory, files, research, images, and agentic behavior. Their economic value now depends more on the surrounding workflow than on a single model benchmark.
Google is the strongest low-price ecosystem bundle. ChatGPT is the broadest general assistant. Claude is the most coherent personal workflow-and-design workbench. Microsoft is strongest when the final artifact must remain inside Office.
Monthly equivalents use published annual prices where available. They do not normalize storage, Office applications, family benefits, promotional pricing, taxes, or regional availability.
400 GB storage. 2× Gemini usage. Pro-model access. Broad Google ecosystem integration.
More messages, uploads, image creation, memory, and context than Free.
$99.99/year or $9.99 monthly. Includes desktop Word, Excel, PowerPoint, Outlook, Copilot, and 1 TB OneDrive.
$200/year or $20 monthly. Includes Code, Cowork, Design, Science, Research, unlimited Projects, and Microsoft 365 integrations.
A Project, Notebook, GPT, Gem, Skill, Agent, and Copilot Page are not interchangeable. The matrix describes practical availability on the lowest personal paid tier.
| Capability | Google AI Plus | ChatGPT Go | Claude Pro | M365 Personal |
|---|---|---|---|---|
| Persistent workspaces | Gemini Notebooks + NotebookLM | Projects | Unlimited Projects | Copilot Notebooks / Pages vary by surface |
| Cross-chat memory | Personalization + activity context | Saved memory + chat history | Memory + past-chat search | Saved memory + chat-history inference |
| Reusable assistant | Gems | Custom GPTs | Skills + Projects | Agents and Notebook instructions |
| Portable skill format | Vendor-specific | Skills rollout is entitlement-dependent | Open Agent Skills format | Vendor-specific |
| Scheduled work | Scheduled actions | Tasks | Remote Cowork schedules | Fragmented; 25 agent tasks ≠ universal scheduling |
| Custom external tools | Google integrations; advanced agent surfaces higher-tier | Apps; custom MCP stronger on organizational tiers | Remote MCP + local extensions | Native Office; broader custom agents favor business |
| Native raster images | Strong | Strong | Structured visuals, not conventional photo generation | Designer/Create with credits |
| Editable design canvas | Canvas | Canvas / Work surfaces | Claude Design | Pages + Office canvases |
| Terminal / IDE workflow | Repository ingestion and coding surfaces | Expanded Codex is a Plus advantage | Claude Code included | Strong Office; developer workflow less central |
| Desktop-native advantage | Web-first; mobile Android integration is material | Work and local integrations expand on desktop | Code, Cowork, Design, local extensions | Full desktop Office applications |
Do not treat “memory” as one feature. A platform may remember user preferences while failing to retrieve a prior project conversation or a source-grounded fact.
Separates saved memories, chat-history inference, Project memory, GPT instructions, and Temporary Chat. Mature personalization, but scope can be misunderstood when global and project context coexist.
Combines memory entries, past-chat search with citations, project summaries, project-specific context, incognito chats, and import/export. It is currently the most inspectable personal memory model.
Personalizes from prior conversations and connected Google services. Notebooks maintain source-and-instruction continuity. Gems and Live may not inherit every personalization behavior.
Separates saved memories, chat-history inference, custom instructions, and the contextual data available inside each Microsoft 365 application. Standalone Copilot and embedded Copilot are different surfaces.
Its primitives divide responsibilities more explicitly: Projects hold context, Skills encode reusable behavior, MCP supplies tools and data, Cowork executes workflows, Design creates visual artifacts, and Code operates on software.
The meaningful differentiator is whether work runs remotely, whether it can use connectors, and whether approvals and failures are observable.
| Platform | Mechanism | Remote execution | Key limitation |
|---|---|---|---|
| ChatGPT Go | Tasks / scheduled prompts | Yes, for supported task types | Tool availability and quotas vary by task and plan. |
| Google AI Plus | Scheduled actions | Yes | Connected-service and feature availability varies by account, region, and surface. |
| Claude Pro | Cowork scheduled tasks | Yes; computer need not remain awake | Shared plan usage and connector security require attention. |
| M365 Personal | Agent tasks and product-specific automation | Depends on agent and application | 25 agent tasks do not imply a general-purpose personal scheduler; advanced agents require Premium. |
A large file can fit while important details remain unretrieved. Visual extraction, context allocation, page count, source format, and retrieval strategy matter more than raw megabytes.
| Platform | Published chat limits | Visual behavior | Practical concern |
|---|---|---|---|
| ChatGPT | 512 MB/file; text documents capped at 2M tokens; images 20 MB; spreadsheets ≈50 MB | PDF visual retrieval depends on plan and surface | A visually rich PDF can be less understood than the file-size limit suggests. |
| Claude | 500 MB/file; up to 20 files/chat; project file limit 30 MB | Text and visuals in PDFs under 100 pages; non-PDF documents generally text-only | Split large visual documents; Project and chat upload constraints differ. |
| Gemini | Up to 10 files/prompt; most files 100 MB; video 2 GB; code repository up to 5,000 files/100 MB | Strong image, audio, video, and repository ingestion | Large-source fidelity improves on higher tiers with larger context and limits. |
| Microsoft Copilot | Up to 20 files/conversation; 50 MB/file | Supports Office, PDF, image, text, and data formats | Long whole-document operations may omit content; localized Office edits are a stronger fit. |
Google and Microsoft have deeper native ecosystems. Claude has the clearest personal path to vendor-neutral external tools. ChatGPT’s strongest managed custom-agent controls remain organizational.
Remote MCP connectors and local desktop extensions support personal tool composition. Plugins can bundle Skills, connectors, and sub-agents. The same openness increases indirect prompt-injection and authorization risk.
Apps connect external services and can combine interfaces, search, actions, and data. Managed internal knowledge, custom MCP policy, approval control, and persistent workspace agents are materially stronger in Business and Enterprise.
Broad consumer integration with Gmail, Calendar, Drive, Docs, Keep, Tasks, Photos, YouTube, Android, and other services. The integration model is powerful but less portable than MCP.
Native depth in Word, Excel, PowerPoint, Outlook, OneDrive, and Microsoft 365 is the primary advantage. Full organizational grounding and Copilot Studio belong to the business architecture.
Raster generation, structured UI design, implementation code, and finished Office deliverables are separate production stages.
Best fit: images, editing, multimodal references, infographics, video, and creative media systems.
Economic edge: these capabilities enter at the lowest subscription price.
Best fit: iterative image generation connected to research, writing, coding, Canvas, and broader artifact workflows.
Tier effect: Plus improves complexity, accuracy, and throughput.
Best fit: editable UI concepts, design systems, SVG, HTML, diagrams, and design-to-code handoff.
Important distinction: structured design rather than conventional photorealistic chat image generation.
Best fit: final Word, Excel, PowerPoint, Designer, Create, Paint, and Photos artifacts.
Personal quota: 60 image credits each month; Premium provides extensive use.
| Output | Primary choice | Reason | Secondary choice |
|---|---|---|---|
| Photorealistic or illustrative image | Gemini | Low-cost multimodal and media breadth | ChatGPT |
| Editable UI concept | Claude Design | Design-system interpretation and implementation handoff | ChatGPT Canvas / Work |
| Research-grounded presentation draft | Gemini Notebook / NotebookLM | Source-corpus grounding | Microsoft PowerPoint agent |
| Production HTML or application | Claude Code | Design + Skills + MCP + code continuity | ChatGPT Codex |
| Finished Office artifact | Microsoft 365 | Native format and application controls | Claude file creation |
| SVG and structured diagrams | Claude | Code-native editable visual output | ChatGPT |
The desktop advantage is largest when local applications, files, IDEs, browser automation, or Office formatting are required. Mobile is strongest for voice, camera, screen context, location, and quick continuation.
Personal products can receive experimental features first. Business products add identity, policy, retention, data controls, auditability, approved connectors, and managed distribution.
Personal plans support GPTs, Projects, Tasks, memory, files, and agentic modes. Business and Enterprise add managed workspace agents, company knowledge, custom MCP governance, administration, and organizational deployment.
Pro exposes unusually broad personal functionality. Team and Enterprise primarily add administration, SSO, role controls, organization-wide Skills, enterprise search, and connector governance.
Google AI consumer plans use personal Google accounts. Workspace Gemini is a separate product and policy surface. A personal feature is not automatically available in a managed Workspace tenant.
Personal and Premium are consumer products. Organizational Graph grounding, Teams and SharePoint context, Copilot Studio, managed agents, and enterprise controls belong to business licensing.
Annual equivalents are used for Claude Pro and Microsoft 365 Personal where noted.
ChatGPT Go for memory, Projects, GPTs, Tasks, files, and images.
Google AI Plus for NotebookLM, Google context, media, and 400 GB storage.
Google AI Plus + ChatGPT Go. Strong research, media, memory, assistants, and cross-checking.
Claude Pro annual + Google AI Plus. Strongest mix of Skills, MCP, Code, Design, Notebook, images, and video.
Claude Pro annual + Microsoft 365 Personal annual. Architect and build in Claude; finish in native Office.
All four lowest paid tiers. Appropriate for standards, migration, portability testing, and comparative evaluations.
The canonical capability package should survive cancellation of any subscription and be deployable into GPTs, Skills, Gems, Notebooks, or Copilot agents.
Markdown, source manifests, glossary, examples, decisions, and provenance.
Skills, workflows, policies, prompts, and deterministic scripts.
MCP, OpenAPI, least-privilege credentials, approvals, and audit events.
GPT, Claude Skill, Gem, Notebook, or Copilot package.
Run identical test cases across platforms and surfaces.
ai-capability-pack/
├── instructions/
├── skills/
├── knowledge/
├── design/
│ ├── tokens.json
│ ├── components.md
│ └── reference-screens/
├── tools/
│ ├── mcp/
│ └── openapi/
├── prompts/
├── evals/
└── adapters/
├── chatgpt/
├── claude/
├── gemini/
└── copilot/
The highest-risk errors come from inconsistent entitlements, false confidence in long context, hidden memory scope, connector injection, shared quotas, and mismatched personal/business identities.
Confirmed direction and informed forecasts are separated below.
Sources were accessed July 20, 2026. Product entitlements, regional prices, quotas, model names, and previews can change without notice. The report favors official product and support documentation.