Which AI Should I Use? Platforms, Models and Best Uses

TL;DR

• There is no single best AI. The right choice depends on the task, the data you will paste in, whether you need sources you can check, and where you already work (Google, Microsoft, a terminal, a browser).
• This guide compares six platforms using only what their own official pages say as of 2026-10-17: ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), Microsoft Copilot, Perplexity, and Grok (xAI). It does not rank them, and it does not use benchmark scores.
• Model names, limits and prices change often. Where I could not read an official page, I say so and tag [VERIFY].
• Whatever you choose: check the data-use setting before pasting anything sensitive, and verify facts that matter.
• Quick decision aid is in section 9; the comparison table is in section 8.

1. How we evaluated (read this first) #

Scope. I picked platforms that are widely known as general-purpose AI assistants. Meta AI and DeepSeek are not covered: I did not research them to the same standard, and the brief was to include them only if sources justified it. Other products exist.

Method. For each platform I read its official pages where I could reach them: model lists, plan and pricing pages, help-centre articles, and privacy documents. I then wrote down what is documented, not what I believe about quality.

What this guide does not do.

• It does not rank platforms. A ranking needs stated criteria, and quality depends on your tasks. If you want to compare, test on your own work (see section 11).
• It does not claim that a feature is missing just because I did not find it. I write "not verified" instead.
• It does not quote benchmark scores or user-review sites.
• It gives prices only where an official page showed them on 2026-10-17, in US dollars, labelled as examples; regional prices, taxes, promotions and plan names differ.

Access limits. Several OpenAI help and pricing pages returned "access denied" to my reader, and Grok's pages did not name models for the consumer app. Where I relied on search-result excerpts of official pages instead of reading the page, I mark it [VERIFY].

Date. Everything below is "as of 2026-10-17". Treat anything older than a couple of months as possibly out of date.

Try these prompts: evaluating tools for yourself #

• create-ai-evaluation-testing-0009: Compare Two Models on My Own Tasks
• create-ai-governance-safety-0006: Vendor Questionnaire for an AI Tool
• create-ai-prompt-testing-improvement-0003: Prompt A/B Comparison Plan
• create-ai-prompt-testing-improvement-0002: Prompt Test Set Builder
• create-ai-workflow-design-0002: Choose Between a Script, a No-Code Flow and an AI Agent
• create-ai-governance-safety-0001: AI Use Policy for a Small Team
• create-ai-governance-safety-0004: Data Handling Rules for Prompts
• create-ai-prompt-testing-improvement-0004: Rubric for Grading AI Outputs

2. ChatGPT (OpenAI) #

What it is. OpenAI's consumer and business assistant, with free and paid plans for individuals and teams, and a separate developer API [2][7].

Current model families (as of 2026-10-17). OpenAI's developer models page lists a flagship group named GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, plus specialised, image, realtime voice, speech and transcription models [2]. A ChatGPT pricing page I read names GPT-6 Luna on Free, Go and Plus, and GPT-6.1 Sol on Plus and above [7]. However, search results for OpenAI help articles used the names "GPT-5.6" and "GPT-6 Pro", so model names in the ChatGPT app may differ from the API list. [VERIFY] in the app's model picker. I will not guess versions.

Documented facts

• The GPT-6 Astra API page lists a context window of about 1.05 million tokens, text and image input, text output, and tools including web search, file search, image generation, code interpreter, computer use and MCP in the Responses API [17]. These are API details; app limits may be lower or different.
• Plans named on the pricing page: Free, Go, Plus, Pro (multiple tiers), Business, Enterprise/Edu [7]. That page is oriented to coding use and shows Plus at $20 per month, Go at $8, and Pro from $100 as of the date read [7]. Treat as examples.
• Image generation is not available on the Free plan according to that page [7].
• Pricing and usage details often come as rolling or weekly limits shared across features [7].

Strengths (documented breadth). A wide model family, with an API offering image, voice and agent tools [2][17]; broad plan range from free to enterprise [7].

Watch-outs. Plan names, model access and limits change often and differ between the app and API [2][7]. Free and low tiers have lower limits [7]. Search excerpts of OpenAI help pages suggest some low-cost tiers may show ads [VERIFY].

Privacy notes. OpenAI's help centre describes an "Improve the model for everyone" setting in Data controls that you can turn off; it applies to new conversations and does not delete existing chats (read via search excerpt; the help page returned access denied) [VERIFY]. Business, Enterprise and Edu plans are described as offering admin and data controls [7]. Read the privacy policy and data-controls page on openai.com before pasting sensitive data.

Use this when: you want a general assistant with a large family of models and broad tool support, or you already use OpenAI's developer platform. Avoid or double-check when: you need a model name to stay fixed (names change), or you handle sensitive data on a plan whose data settings you have not confirmed.

3. Claude (Anthropic) #

What it is. Anthropic's assistant (web, desktop, mobile) and developer platform. It also makes Claude Code, a terminal coding agent, available on paid plans [5].

Current model families (as of 2026-10-17). Anthropic's models page lists Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5 as the current lineup, each with a 1 million token context window and 128K maximum output on the API, text and image input, text output, and tool use [1]. It describes Fable 5.1 as for demanding reasoning and long-horizon agentic work, Opus 5.5 for long-running agentic coding and knowledge work, Sonnet 5.5 as a speed and intelligence balance, and Haiku 5.5 for high-volume, latency-sensitive tasks [1]. Older models are listed as legacy [1].

Documented facts about the consumer plans [5]

• Plans: Free, Pro, Max (two usage tiers), Team, Enterprise, Education.
• Free includes Sonnet and Haiku models; Pro and higher add Opus; Fable access is tied to usage credits or higher tiers [5].
• Every plan has usage limits that reset on a rolling five-hour window; paid plans also have weekly limits [5].
• Claude Code is included in paid plans and shares the same usage pool [5].
• Research is available on paid plans; projects, artifacts, connectors and skills are listed for various tiers [5].
• Example list prices shown on 2026-10-17: Pro $20 per month if billed monthly ($17 per month billed annually); Max from $100 per month [5].
• Team and Enterprise plans are described with admin controls and no model training on content by default [5].

Strengths (documented). Very long context on current models [1]; a published model-selection page and model IDs; Claude Code and an Agent SDK for coding and agent work [5].

Watch-outs. Usage limits are explicit and can interrupt long sessions [5]. The consumer app's image generation, voice and other features were not verified from official pages here. [VERIFY] on the product page.

Privacy notes. Anthropic's privacy policy (effective 2026-09-10) says it may use inputs and outputs to train and improve models and that you can opt out in account settings; flagged conversations and reported material may still be used; deleted conversations are removed from history immediately and from back-end systems within 30 days [14]. Team and Enterprise plans are described as not training on content by default [5].

Use this when: you work with long documents or large codebases, want a terminal coding agent, or want a documented model lineup with stated context sizes [1][5]. Avoid or double-check when: you cannot tolerate usage caps, or your data must not be used for training on a consumer plan without first changing the setting [14].

4. Gemini (Google) #

What it is. Google's assistant (Gemini app and Google integrations) with a developer API and AI Studio [3][6].

Current model families (as of 2026-10-17). The Gemini API models page (last updated 2026-10-13) lists Gemini 4 Argon as the flagship reasoning model with a 1 million token context window, plus Gemini 3.8 Flash, Flash-Lite and Live variants, Gemini 3.x models, image models (Nano Banana), video (Veo, Gemini Omni Flash), music (Lyria), Deep Research agents, an Antigravity Agent, embeddings and robotics models [3]. The page does not state context windows for most other models [3].

Documented facts about consumer plans [6]

• Tiers: Free, Google AI Plus, Google AI Pro, Google AI Ultra; business tiers under Gemini Enterprise.
• Free includes a Flash-Lite model, image generation and editing, Deep Research, Gemini Live, Canvas, Gems, and 15 GB storage as listed [6].
• Higher tiers list higher usage limits, video generation, more storage, and access to the flagship Gemini 4 Argon [6].
• Example list prices on 2026-10-13 pages: AI Plus $4.99 per month, AI Pro $19.99 per month, AI Ultra from $99.99 per month [6].
• The Gemini app can connect to other Google apps (for example a mention such as @Messages on Android) and supports file uploads, notebooks, schedules and more, per Google help pages [18][20].

Strengths (documented). Close integration with Google products and storage; a broad model catalogue covering text, image, video and audio on the developer side [3][6].

Watch-outs. Features and models vary by plan, region and device; Google help notes that some app extensions are not available on every surface [20]. Model names and tiers are changing quickly (see "Preview" and "legacy" labels on the model page) [3].

Privacy notes. Google's Gemini Apps Privacy Hub says the Keep Activity setting controls whether chats are saved and used to improve Google services, including AI models; when on, a subset of chats may be reviewed by trained human reviewers after being disconnected from the account and kept for up to three years; activity auto-deletes after 18 months by default (adjustable); temporary chats are excluded from improvement; some chats are kept up to 72 hours for service and safety even with the setting off [15].

Use this when: you live in Google Workspace and Search, want image/video/audio generation in the same ecosystem, or use Google's developer tools [3][6]. Avoid or double-check when: you need to be sure about human review and retention; read the Privacy Hub and set Keep Activity deliberately [15].

5. Microsoft Copilot #

What it is. Microsoft's family of assistants: a consumer Copilot app and Microsoft 365 apps for individuals, and a work version (Microsoft Copilot, previously called Microsoft 365 Copilot) that uses your organisation's Microsoft 365 data [8][16].

Current model families. Microsoft's individual-plans page does not name specific models [8]. The Microsoft privacy documentation says Copilot orchestrates large language models with content in Microsoft Graph and Microsoft 365 apps, and that Microsoft offers third-party models such as Anthropic and OpenAI models in Copilot experiences; admins can choose whether to use them, and Microsoft may deploy other models it hosts [16]. Do not assume a particular model: [VERIFY] in your tenant's documentation.

Documented facts

• Copilot works inside Word, Excel, PowerPoint, Outlook and Teams for drafting, summarising and analysis; a standalone Copilot app offers chat; named features include Researcher and Analyst, AI Restyle for photos, and "Cowork" for multi-step tasks [8].
• Individual plans listed on 2026-10-17: Microsoft 365 Personal ($9.99/month), Family ($12.99/month), Premium ($19.99/month), Pro ($99.99/month, monthly billing only) with different Copilot usage levels [8].
• In Copilot Chat, typing / opens a menu to reference files and, with the right licence and setting, people, meetings, emails and chats [19].
• Excel features require files saved to OneDrive with AutoSave on [8].

Strengths (documented). Native integration with Microsoft 365 apps and work data, with access limited to what each user is permitted to see [16][19].

Watch-outs. Features depend on licence and admin settings (for example Work IQ, agents, web search); the product and plan names have changed more than once, and Microsoft notes the rename in its own docs [16]. Third-party models may have different data-boundary terms (for example Anthropic models are noted as excluded from the EU Data Boundary in Microsoft's documentation) [16].

Privacy notes (work version). Microsoft states that prompts, responses and data accessed through Microsoft Graph are not used to train foundation models, that Copilot only surfaces content a user already has permission to view, that interactions are stored and manageable by admins (for example via Microsoft Purview), and that users can delete their Copilot activity history [16]. Consumer Copilot terms are a separate matter: read the Microsoft privacy statement for the app you use. [VERIFY]

Use this when: your documents, email, meetings and chats live in Microsoft 365 and your organisation licenses Copilot [16]. Avoid or double-check when: you do not know which version (consumer or work) you are using, or which data it can see.

6. Perplexity #

What it is. An AI-powered search and answer service that shows citations; Pro adds more search depth, research mode, file uploads, and more [9].

Current model families (as of a Perplexity table dated 2026-09-04). Perplexity lists its own Sonar 2 and models from several providers: OpenAI (GPT-5.6 Terra, GPT-5.6 Sol), Google (Gemini 3.7 Flash), Anthropic (Claude Sonnet 5, Claude Opus 5), Moonshot AI (Kimi K3), Z.ai (GLM 5.3), xAI (Grok 4.6), NVIDIA (Nemotron 3 Ultra). Pro includes all except GPT-5.6 Sol and Claude Opus 5; Max includes all; Free has limited Pro Searches and no manual model choice. The page says models are subject to change and the in-app selector is the source of truth [10].

Documented facts. Pro includes advanced models from several providers, a higher daily volume of Pro Search, research mode that produces reports, uploads of documents, spreadsheets, audio, video and images, roughly ten times as many sources per answer, creation of files and simple apps, and some media generation [9]. Access to some models may be limited during heavy demand [9].

Strengths (documented). Citation-forward answers and multi-provider model choice in one place [9][10].

Watch-outs. Model availability differs by plan and may be limited at peak times [9][10]. As an aggregator it relies on third-party models, so behaviour can change when providers change.

Privacy notes. For Free, Pro and Max users, the "AI data retention" setting is on by default, allowing data to be used for AI training; you can turn it off in Account preferences; opt-outs apply only to data collected afterwards. Enterprise data is stated to be excluded from training. Perplexity does not specify consumer retention periods on that page [11].

Use this when: your main job is finding and checking sources quickly, or you want to try several providers' models without separate accounts [9][10]. Avoid or double-check when: you need a fixed, named model for repeatable results, or you have not changed the AI data retention setting and the content is sensitive [11].

7. Grok (xAI) #

What it is. xAI's assistant on grok.com and iOS/Android apps, with image and video creation (Grok Imagine), voice, file uploads, and connectors to email, files and calendar apps [12].

Current model families. xAI's developer models page lists text models such as grok-4.7 (described as the most capable, with a 500k context window and a May 2026 knowledge cutoff), grok-4.6, grok-4.5, grok-4.3, grok-4.20 variants (1M context for the 4.20 and 4.3 entries), and grok-build-0.1; plus image, video and voice models [4]. The consumer overview page names no specific models and does not mention web search, so which model powers the app is [VERIFY] [12].

Documented facts. Grok is free to start; paid plans raise limits and draw from a single weekly usage pool [12]. xAI's "X Subscription" page lists Free, Plus, Premium, Super and Ultra tiers; example list prices shown: Plus $8, Premium $30, Super $100, Ultra $200 per month, with Premium and higher including "Grok Bot" (persistent cloud-computer agent teammates) [13][12]. Business and Enterprise team plans are mentioned [12].

Strengths (documented). Combined chat, image/video, voice and agent-style features in one subscription, with a published API model list and context sizes [4][12][13].

Watch-outs. Plan structure has been changing (older SuperGrok plans continue while a new bundle exists) [13]; the consumer page does not state models or search. I did not find an official privacy statement I could read for training use, so I give none. [VERIFY] with xAI's privacy policy before pasting sensitive data.

Use this when: you want an assistant bundled with X and image/video tools, or xAI's API. Avoid or double-check when: you need documented, verifiable data-use terms before sharing sensitive content (not verified here).

8. Comparison table by task (documented features only) #

This table says what the official pages document for each use case. "Documented" is not the same as "best". "Not verified" means I did not find or could not read an official statement.

Try these prompts: matching a tool to a task #

• create-ai-workflow-design-0002: Choose Between a Script, a No-Code Flow and an AI Agent
• create-ai-evaluation-testing-0009: Compare Two Models on My Own Tasks
• create-ai-agent-prompts-tools-0003: Research Agent Instructions With Source Rules
• create-ai-custom-instructions-assistants-0008: Research Assistant Instructions With Citations
• create-spreadsheets-formulas-functions-0032: Text Cleaning Formulas for Messy Data
• create-spreadsheets-excel-practice-0008: Excel PivotTable Summary From a Raw Table With Cross-Checks
• audit-legal-contracts-0001: Contract Plain-Language Risk Summary
• create-spreadsheets-project-tracking-0013: Meeting Action Item Tracker
• create-ai-governance-safety-0006: Vendor Questionnaire for an AI Tool
• create-ai-governance-safety-0004: Data Handling Rules for Prompts

9. Decision guide: if you need X, start with Y (then test) #

This is a starting point, not a recommendation of quality. Always run your own test (section 11).

10. Privacy and data: a quick comparison #

Rules that apply everywhere: do not paste passwords, keys or other people's private data; read the setting names (they differ); use temporary or incognito modes where offered; and ask your organisation which tools are approved.

11. How to run your own fair test (30 minutes) #

1. Choose 5 real tasks you do weekly (non-sensitive versions).
2. Write one prompt per task and use the same prompt in each tool.
3. Score outputs 1 to 5 on accuracy, completeness, tone, format and usefulness, with a short reason. Blind-label outputs if you can.
4. Check any facts and quotes against sources.
5. Note speed, limits hit, and ease of use.
6. Repeat each prompt twice to see consistency.
7. Pick the tool for each task, not one tool for everything. Re-test when plans or models change.

12. Common mistakes #

13. Checklist before you commit #

• [ ] I know my top three tasks and what "good" looks like for each.
• [ ] I tested at least two tools on the same real tasks.
• [ ] I checked the plan page for limits and model access today.
• [ ] I read the data-use setting and set it deliberately.
• [ ] I know which of my data must never be pasted.
• [ ] I verified any facts the output will be relied on for.
• [ ] I recorded the date and tool version I tested.

14. Glossary #

• Model family: a group of related AI models offered under one brand name.
• Context window: how much text a model can consider at once (counted in tokens).
• Token: a chunk of text used to measure input and output size.
• Deep research / research mode: a feature that runs several searches and writes a longer, sourced report.
• Agent: an AI system that plans steps and uses tools (see Guide 2).
• MCP: Model Context Protocol, a standard for connecting tools and data to AI apps.
• Connector: a link that lets an assistant read data from another app.
• Opt-out: a setting that stops your content being used to improve models.
• Pinned model ID: a fixed identifier for a specific model snapshot.
• Tenant: an organisation's own Microsoft 365 environment.

Sources #

All accessed 2026-10-17. Pages marked "partly read" were only partly retrieved. Where the page showed no date, "none shown".

1. Models overview. Anthropic (Claude Platform docs). https://platform.claude.com/docs/en/about-claude/models/overview (none shown)
2. Models. OpenAI Developers. https://developers.openai.com/api/docs/models (none shown)
3. Models. Gemini API, Google AI for Developers. https://ai.google.dev/gemini-api/docs/models (last updated 2026-10-13)
4. Models. xAI docs. https://docs.x.ai/docs/models (none shown)
5. Plans and pricing. Claude (Anthropic). https://claude.com/pricing (none shown)
6. Gemini subscriptions. Google. https://gemini.google/subscriptions/ (none shown)
7. Pricing. ChatGPT Learn, OpenAI. https://learn.chatgpt.com/docs/pricing (none shown; coding-oriented)
8. Microsoft Copilot for individuals. Microsoft. https://www.microsoft.com/en-us/microsoft-copilot/for-individuals (none shown)
9. What is Perplexity Pro? Perplexity Help Center. https://www.perplexity.ai/help-center/en/articles/10352901-what-is-perplexity-pro (none shown)
10. What advanced AI models are included in my subscription? Perplexity Help Center. https://www.perplexity.ai/help-center/en/articles/10354919-what-advanced-ai-models-are-included-in-my-subscription (table as of 2026-09-04)
11. Data Collection at Perplexity. Perplexity Help Center. https://www.perplexity.ai/help-center/en/articles/11564572-data-collection-at-perplexity (none shown)
12. Welcome to Grok. xAI docs. https://docs.x.ai/grok/overview (none shown)
13. Pricing (The X Subscription). xAI docs. https://docs.x.ai/x-subscription/pricing (none shown)
14. Privacy Policy. Anthropic. https://www.anthropic.com/legal/privacy (effective 2026-09-10)
15. Gemini Apps Privacy Hub. Google Gemini Apps Help. https://support.google.com/gemini/answer/13594961 (last updated 2026-10-13)
16. Data, Privacy, and Security for Microsoft Copilot. Microsoft Learn. https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy (ms.date 2026-07-09; updated 2026-10-08)
17. GPT-6 Astra model page. OpenAI Developers. https://developers.openai.com/api/docs/models/gpt-6-astra (none shown)
18. What you need to sign in to Gemini Apps. Gemini Apps Help, Google. https://support.google.com/gemini/answer/13278668 (none shown)
19. Refer to specific files and more in Microsoft Copilot. Microsoft Support. https://support.microsoft.com/en-US/Microsoft-365-Copilot/refer-to-specific-files-and-more-in-microsoft-365-copilot (none shown)
20. Send messages with the Gemini mobile app on Android. Gemini Apps Help, Google. https://support.google.com/gemini/answer/15574940 (none shown)

OpenAI help-centre statements marked [VERIFY] come from search-result excerpts of official pages that my reader could not open; they are not counted in the 20.

Research log: ops/research/guide-4-which-ai.md.

Last updated: 2026-10-17