Executive summary
This document assesses the LLM applications suitable for individual productivity.
Recommended: Claude Team. It ranks first on both axes assessed: power — the app, the harness and the models combined — and adoption by a non-technical user. It reaches Microsoft 365 through connectors: reads mail, calendar, files and Teams, writes to mail, calendar and files.
Assessment
Power: what can be achieved with the app, the harness and the models together.
Adoption: ease of adoption for a non-technical user — building workflows, organising work, templates, training.
Integration: native, or through connectors authorised in the tenant.
Assessment table
| Anthropic | Microsoft | OpenAI | xAI | ||
|---|---|---|---|---|---|
| Plan | Claude Team | Gemini Enterprise Business | Microsoft Copilot | ChatGPT Business | SuperGrok Business |
| Models | Opus 5 · Sonnet 5 · Haiku 4.5. Fable runs on credits on a standard seat; only a Premium seat carries it inside the limit | Gemini 3.7 Flash and the rest of the Gemini models | The seat names none. Its picker offers Auto, Quick response, Think deeper. The named roster — GPT 6 Astra · GPT 5.6 Sol · GPT 5.6 Terra · Claude Opus 5 · Claude Sonnet 5 · Claude Fable 5.1 — belongs to Copilot Cowork, which is paid above the seat | GPT-5.6 Sol · Sol Pro · Terra · Luna · GPT-5 Thinking Mini | All Grok models |
| Document handling | Read whole into context. 200k or 1M tokens — Anthropic's own pages state both. PDF to 1,000 pages, images read to page 100. Projects switch to retrieval near the window | Indexed and retrieved. No context window published for this edition; 25 GiB of indexing per seat | Indexed and retrieved — Microsoft's Retrieval API returns "relevant text chunks". No context window published. 512 MB per file, five files per action in OneDrive | Read whole into context. 54K and 256K on the pricing page, 128K and 272K on the models page — OpenAI's own pages state both. Images are discarded from uploads | 500k token context on Grok 4.6. 150 MB a file, about 100 files on web |
| Coding agent | Claude Code, in the seat. Same usage pool as the app | Not included. Gemini Code Assist is Standard or Plus, or $19 a seat standalone | Not included. GitHub Copilot is a separate product, $19 a seat | Codex, in the seat. Shared usage pool | Grok Build, in the seat. Shared usage pool |
| Ready to use templates | Industry packages — financial services, legal, life sciences, small business — connectors, skills and agent templates | Deep Research, Gemini Notebook, Data Insights. Partner agents and an Agent Designer | Sales, Service and Finance agents. Researcher, Analyst, Facilitator. An Agent Store | None found. Apps and plugins, built by the customer | Solution pages only. No package found |
| Training material | Claude Academy — courses, quizzes, webinars, completion badges | Skillshop courses and a resources hub | Success Kit, Copilot Academy, Microsoft Learn paths | OpenAI Academy — three course tracks and live events | Product guide only; no programme found |
| Training in Spanish | Yes | Not found on Skillshop | Yes | — | — |
| MS 365 integration | Add-ins for Excel, Word and PowerPoint; Outlook in beta. Reads SharePoint, OneDrive, Outlook and Teams. Writes to mail, calendar and files. Anthropic's own pages disagree on writing to Teams: the July release note says read-only, the current connector article lists sending Teams messages | No add-ins. Reads SharePoint, OneDrive, Outlook and Teams. Write actions are generally available: upload to SharePoint and OneDrive, send mail, message Teams | Native | Add-ins for Excel and PowerPoint. Reads and writes SharePoint, OneDrive, Outlook and Teams. Calendar is read and write — create, update, RSVP, cancel | No add-ins. Connectors read and write SharePoint, OneDrive, Outlook and Teams. SharePoint and OneDrive need the Business tier |
| Single sign-on | Yes, SAML 2.0. Just-in-time provisioning; no SCIM | — | Yes, through the Microsoft 365 accounts the company already has | Yes, SAML and OIDC, enforceable; no SCIM | — |
| Audit logs | No. Usage analytics only — connector read and write counts, file operations, spend export | — | Yes, Purview. Prompts, responses and file access, 180 days | Admin activity only; conversation content is Enterprise | Yes, a team log of API and admin activity |
| Per user, annual / monthly | $20 / $25 standard · $100 / $125 premium | $21 / not published | $30 / — | $20 / $25 standard · $100 / $125 premium | Not published |
| Paid above the seat | Optional credits past the limit | None found | Yes. Cowork agents and workflows are not included in the seat. $0.01 per credit | Optional credits past the limit | Optional credits past the limit |
Sourced to each vendor's own pages, read 21 and 22-Aug-2026.
Product comparison
| Power | Adoption | |
|---|---|---|
| Anthropic · Claude Team | Frontier models. A document is read whole in chat, images included to page 100 of a PDF. One skill runs in the app, in the Office add-ins and in Claude Code | Workflows built by describing them, or by recording yourself doing one. Projects hold documents and instructions. Templates for the file types it produces. Courses, quizzes and four certifications, in English |
| Google · Gemini Enterprise Business | Gemini 3.7 Flash and the rest. Documents are indexed and retrieved, not read whole; no context window is published for this edition. Gems hold instructions; they do not run workflows | Four steps from Agents to a working agent in Agent Designer. Deep Research is included. Gems belong to the consumer Gemini app, not to this edition. No course found on Skillshop |
| Microsoft · Copilot | Frontier models, bought from OpenAI and Anthropic, but the seat neither names nor lets the user select one; the named roster sits in Cowork, billed on credits. Documents are indexed and retrieved, not read whole, and no context window is published. Acting on a service outside Microsoft 365 needs Copilot Studio, a separate product with its own billing | Copilot Notebooks hold up to 300 files with standing instructions. Agent Builder configures an agent from eight named fields; an agent that acts on an external service is outside the seat. 21 agent templates. Training material published in Spanish |
| OpenAI · ChatGPT Business | Frontier models. Images are discarded from uploads, so a scan returns text only. Custom GPTs run in the ChatGPT app only; anything beyond is the API | Describe what you want and ChatGPT drafts it. Projects hold documents. No templates and no worked example published. Three course tracks, in English |
| xAI · Grok Business | All Grok models. Five built-in skills produce documents, spreadsheets, decks and PDFs. No context window, document limit or build procedure appears on any Business page | Skills created by describing them in conversation; no build procedure is published. Projects and Collections hold document sets, built around data rooms. No training material |
Claude Team — Details
Included. Frontier models. A document read whole in chat. Claude Code in the seat — terminal and the main code editors, on the same usage pool as the app. One skill runs in the app, in the Office add-ins and in Claude Code. Projects. Single sign-on.
Not included.
- Audit logs. Enterprise only. Team sees usage counts, with no per-user trail.
- Writing to Teams. Anthropic's own pages disagree: the July release note says read-only, the current connector article lists sending Teams messages.
- Scoping SharePoint. All sites the user can already see, or none.
- Fable on a standard seat. It runs on usage credits; only a Premium seat carries it inside the limit.
Annex
Not compared
Grok Bot, launched 11-Aug-2026 in beta, is sold to SuperGrok Plus and Heavy and to Cursor tiers. Enterprise is no longer waitlist-only, and it is sold through Cursor's plan structure rather than xAI's; Business is not listed. Each bot runs on its own cloud machine with its own logins.