For most finance professionals paying for one plan, Claude Pro is the first $20 to spend. It includes Claude for Excel, and every current Claude model reads up to 1M tokens in chat. ChatGPT Plus, also $20, is the better first plan when your week is a broad mix of drafting, research, images, and voice. Google AI Pro, at $19.99, comes first only when your working life runs on a personal Google account, because Google sells it to personal accounts only; a firm on Google Workspace gets Gemini through its Workspace edition instead. None of the three is the right place for client data: that needs a business plan. Plans and prices below were checked on the vendors' pages on 31 October 2026, and feature details on 22 October 2026.

For which model suits which finance job, see ChatGPT vs Claude vs Gemini; this post answers the buying question.

Three plans at roughly $20: what each one buys

US dollars a month, before tax. Anthropic publishes dollar prices only. OpenAI bills UK subscribers in sterling but shows those prices only after sign-in, so no sterling price is given for either.

ChatGPT PlusClaude ProGoogle AI Pro
Price$20, monthly only; billed in local currency in the UK, so check the pricing page$20 monthly, or $17 a month billed annually ($200 up front)$19.99 (£18.99 in the UK)
ModelsGPT-6 in Chat (rolling out) at Instant, Medium, and High thinking; GPT-6.1 Sol in Work; limited GPT-6 Astra in WorkOpus, Sonnet, and Haiku; Fable 5.1 on usage creditsGemini 4 Argon, plus Google's Pro and Flash models
Long documentsNo Chat context window published for Plus; 500,000 tokens in WorkUp to 1M tokens in chat on current models1M tokens, which Google puts at up to 1,500 pages
Spreadsheet add-inChatGPT for Excel (also on Free); ChatGPT for Google Sheets is a separate add-onClaude for Excel, generally available; Claude for Google Workspace in public betaGemini in Google Sheets, Docs, and Gmail
Deep researchIncluded "(where available)"Research, on every paid planDeep Research is on every tier; AI Pro writes reports with the Pro model and higher limits
AgentChatGPT WorkTasks in any conversation (rolling out), and Claude CodeTasks in Gemini, not in the UK, the EEA, Switzerland, or Nigeria
Training on personal plansOn unless you switch it offAnthropic's label: "Opt-out"Keep Activity on by default; chats can train models and a subset is read by human reviewers

The add-in and long-document rows decide most finance purchases; the training row decides what goes into any of them.

Spreadsheets and file analysis

All three analyse an uploaded workbook by writing and running code, and all three hand back files. The differences are in the limits and the cautions.

  • ChatGPT Plus runs data analysis in Python "in a stateful Jupyter notebook environment" that cannot make external web requests, and OpenAI warns that image-based or scanned tables may not extract reliably. Uploads go up to 512 MB a file, with spreadsheets at about 50 MB and up to 80 files every three hours.
  • Claude Pro needs its code execution and file creation switch, which is on by default on Pro, for .xlsx uploads, file creation, and analysis. Keep each file under 30MB, which fits every limit Anthropic publishes; a chat takes up to 20 files and PDFs of up to 1,000 pages.
  • Google AI Pro accepts documents and spreadsheets of up to 100 MB a file, and Gemini can hand back a finished Google Sheet or an Excel or CSV file from the chat.

Excel and Sheets: where the add-ins run

The add-in matters more than the chat window, because it works on the file the review happens in.

Claude Pro includes Claude for Excel, PowerPoint, and Word. In Excel it answers with cell-level citations, changes assumptions while keeping formula links intact, traces errors to their source, and can pull data through connectors such as S&P Global, LSEG, and Daloopa. The Corporate Finance Institute ran a published test in which the add-in rebuilt the deleted forecast section of a three-statement model, and its year 5 net income, forecast balance sheet, and ending cash matched CFI's own completed model. CFI sells the course the model comes from, so read it as a supplier's test; AI for month-end close walks through it. Anthropic itself does not recommend it for audit-critical calculations without verification, and it does not support macros and VBA, data tables, Excel 2016 or 2019 perpetual licences, or Excel on iPad or Android.

ChatGPT Plus has ChatGPT for Excel, one Microsoft add-in that also covers PowerPoint and Word. It is on every plan, Free included, so a team can try it before paying. Its default skills cover financial modelling and corporate-finance formatting, its spreadsheet chats are kept apart from your ChatGPT history and memory, and VBA and macros may not be fully supported. On Plus it shares the Work and Codex allowance.

Google AI Pro puts Gemini inside Google Sheets on a personal account. The =AI("prompt", [range]) function generates results for only the first 350 selected cells, so a long ledger needs splitting, and Sheets canvas turns a sheet into an interactive dashboard on AI Pro.

Deep research: a report, then the checking

All three run a long research job that reads web sources and writes a report. The reviewer's job is unchanged: open the source behind every figure and quote before the report goes anywhere.

  • ChatGPT Plus proposes a research plan you can edit, lets you interrupt and refocus mid-run, and can restrict research to, or prioritise, sites you name. Point it at the standard-setter or the regulator. Reports export as Markdown, Word, or PDF.
  • Claude Pro runs Research across the web and your connected tools, including Gmail, Google Calendar, and Google Docs. It runs on Opus 5.5 (falling back to Opus 4.8), whichever model the chat is set to, and it uses your limits faster than ordinary chat.
  • Google AI Pro uses Google Search as its default source, lets you upload files and add notebooks, and reaches Gmail and Drive only once the Google Workspace app is connected to Gemini.

Agents and tasks: what each will do on its own

Each plan includes an agent that works through a multi-step job. What matters is what it can touch and when it stops to ask.

ChatGPT Work produces a document, spreadsheet, presentation, report, or Site. Its browser runs on a separate cloud computer and confirms before payments or bookings; keep site access on "Always ask". Work is not available in a project set to project-only memory, so an isolated client project gives it up.

Claude Pro is rolling out tasks inside any conversation. Since 6 October 2026, new tasks on Pro run in Anthropic's cloud, and a file a task needs from your computer is copied to Anthropic's servers. For work that must stay on your machine, Anthropic points to Claude Code in the desktop app, which Pro includes.

Tasks in Gemini needs AI Pro or Ultra, a personal Google account, and Keep Activity switched on, and it is not available in the UK, the EEA, Switzerland, or Nigeria. For a UK reader, Google AI Pro has no agent.

Usage limits: what runs out first

  • ChatGPT Plus has no published thinking quota in Chat. In Work, OpenAI estimates about 15 to 160 messages per 5-hour window on GPT-6.1 Sol, with only limited GPT-6 Astra. Plus stops at High thinking: Extra High and the Pro level need a Pro plan, from $100, or a Business or Enterprise seat.
  • Claude Pro resets on a rolling 5-hour window with a weekly limit on top, and chat, the apps, and Claude Code share one allowance, at least 5 times Free's per session. Tasks that search, run code, or create files use it faster.
  • Google AI Pro measures compute, not messages: the allowance refreshes every 5 hours up to a weekly limit, at 4 times the free tier. Deep Research and the Pro model use more.

Training and privacy defaults on personal plans

On all three personal plans, the default lets the vendor use what you type. Change the setting before any work content goes in.

Setting to changeWhat still applies
ChatGPT PlusSwitch off "Improve the model for everyone" in Settings > Data controlsA thumbs up or down lets OpenAI train on the whole conversation, even after you opt out; Codex has its own separate setting
Claude ProSwitch off "Help Improve our AI models" in Settings > PrivacyRating a response sends the whole conversation to Anthropic, kept for up to 5 years and usable for training
Google AI ProSwitch off Keep Activity, or at least set auto-delete to 3 monthsPaying for AI Pro does not change Google's consumer terms; with Keep Activity off, most connected apps such as Gmail and Drive stop working

The Google row is a real trade-off: Gemini's Gmail and Drive features depend on the setting that lets Google train on your chats. Google's own privacy hub asks users not to enter confidential information they would not want a reviewer to see.

When the answer is a business plan instead

If the work involves client files, payroll, or anything under a confidentiality obligation, the first purchase is not one of these three. The business plans change the default: ChatGPT Business, Claude Team, and the core Google Workspace editions do not train on your data by default, and they add admin controls; ChatGPT Business and Claude Team also offer single sign-on.

  • ChatGPT Business: $25 a seat a month, or $20 billed annually, for a Standard seat; at least two seats.
  • Claude Team: $25 a seat a month, or $20 billed annually, for a Standard seat; two to 150 seats.
  • Google Workspace Business Standard: $14 a user a month on an annual commitment, with Gemini in Docs, Sheets, and Drive; Business Starter at $7 has Gemini in Gmail but not in Docs or Sheets.

A personal plan still works for published or anonymised material. Client names and live figures stay out until the firm pays for a plan that does not train on them.

Which to buy first: the decision table

The AI Playbook maps all three platforms job by job, with 14 copy-paste prompts; the table below covers only the buying decision.

Your situationBuy firstWhyCheck before you pay
Firm on Microsoft 365, work lives in ExcelClaude ProClaude for Excel, with cell-level citationsWhether IT allows AppSource add-ins; try ChatGPT for Excel on Free first
Firm on Google WorkspaceYour Workspace edition, then Claude Pro or ChatGPT PlusGoogle AI Pro is personal-only; Business Standard and above has Gemini in Docs and SheetsYour firm's edition; Claude for Google Workspace is in beta
A personal Google account runs your working weekGoogle AI ProGemini in Gmail, Docs, and Sheets, a 1M-token window, and 5 TB of storageKeep Activity is on by default; no Tasks in the UK
Long documents and contractsClaude ProUp to 1M tokens in chat, roughly 555,000 words, and PDFs of up to 1,000 pagesOlder models read less; check the model picker
Broad mixed workChatGPT PlusWork, deep research, images, and voice in one subscriptionPlus stops at High thinking; Work allowance on Plus is limited
Client or confidential dataA business planNo training by default, plus admin controlsMinimum two seats on ChatGPT Business and Claude Team

Plan on two in time: buy for the most painful recurring task first, and add a second only when a task proves it needs one.

A one-week test before you commit

Take each plan monthly, and test the two that fit your situation (all three cost about $60). Use last month's signed-off work, names removed, so you already know the answers.

Day 1. Settings. Switch off the training settings above before anything is uploaded, and write down the answer to each test.

Day 2. A long document. Give each the same agreement or report of 100 pages or more that you have already reviewed. Ask for the 10 clauses or figures that matter, with page references, and score them against your own notes.

Day 3. A workbook. Run last month's management accounts through each add-in and ask for variance commentary. Tie every figure in the draft to a cell. If more than two figures fail, that plan fails the test.

Day 4. Deep research. Ask each a question you have already researched, such as a recent standard or regulatory change. Open every source behind every figure and count the errors.

Day 5. A task. Ask each agent for one finished deliverable from a folder of test files, with approvals switched on. Note what it asked before acting and what it did without asking.

Day 6. A heavy day. Run your heaviest real day of non-confidential work and note when each plan hit a limit.

Day 7. Score and cancel. Compare errors found, minutes of rework, and limit hits. Keep the plan that won on the task you do most often, and cancel the rest before the month renews.

The test leaves a record of which plan, which task, what failed, and who checked it: the evidence a reviewer will ask for once the tool reaches live work.

Where to go next

Start with The AI Playbook, free with your email address: 51 pages, a 45-minute read, comparing ChatGPT, Claude, and Gemini, with 14 copy-paste prompts. Once you have chosen a plan, The Claude Playbook, The ChatGPT Playbook, and The Google AI Playbook take that platform into professional finance work in depth, at $47 each, paid once. If you expect to run two or three of them, the Complete Collection has all three for $97, a $44 saving.

Sources

Plans and prices checked on 31 October 2026; feature details checked on 22 October 2026.