Aidar Akmaev
Disclosure: NOXI publishes this ranking. NOXI is a presentation maker, not an AI chatbot, so it is not one of the nine scored assistants; it appears only in the section for readers whose goal is a presentation and in the closing build-it flow. The rubric, the weights, every rating and every official source are printed below so that you can re-score any assistant yourself.
Quick answer: Pick Google Gemini (94 of 100) for the widest documented free plan, a $4.99 entry plan and a 1M-token window on AI Pro. Pick ChatGPT (88) for a full free plan, five tools on Plus and six surfaces. Pick Claude (85) for a published 1M-token window and custom connectors even on Free. Official pages, checked 22 October 2026.
No vendor publishes a measure of answer quality. Vendors do publish what the free plan includes, what a paid plan costs, how long a document it reads, which tools it has, what connects to it, whether chats train its models and where it runs. This page scores nine assistants on those seven facts, one printed rule each, with the arithmetic.
In short
- RankingGoogle Gemini 94, ChatGPT 88, Claude 85, Grok 83, Mistral Vibe 79, Microsoft Copilot 77, Perplexity 65, Meta AI 64, DeepSeek 39.
- Second placeChatGPT leads Claude 88 to 85 on free plan, price and tools; Claude leads on long documents and add-ons.
- MethodDocumented capability from official pages opened on 22 October 2026, not a new hands-on test. Answer quality is not scored.
- Most used44% of US adults say they ever use ChatGPT, 24% Gemini and 17% Copilot (Pew Research Center, survey February 2026).
- Best free plansGemini, ChatGPT and Mistral Vibe document all five free-plan capabilities we count.
- No-training statementMicrosoft states that chats in the new Copilot app are not used to train its foundation models.
Key numbers
The ranking: 9 AI chatbots scored on 7 rules
Google Gemini leads with 94 of 100, ChatGPT follows at 88, Claude at 85 and Grok at 83; the full order is in the table, with a "best when" line that rests on a documented fact for each assistant. The same order answers "best AI assistant", since vendors use both words for these products.
| # | Assistant | Score | Best when |
|---|---|---|---|
| 1 | Google Gemini | 94 | You want the most documented features per dollar, from $4.99 to a 1M-token window at $19.99. |
| 2 | ChatGPT | 88 | You want a full free plan, five tools on Plus and the assistant on six surfaces. |
| 3 | Claude | 85 | You read long documents (up to 1M tokens) and want your own connectors, one even on Free. |
| 4 | Grok | 83 | You want custom connectors for every user and a paid tier from $8 a month. |
| 5 | Mistral Vibe | 79 | You want a full free plan with PowerPoint export and custom MCP connectors. |
| 6 | Microsoft Copilot | 77 | A stated no-training rule matters most, or you want Copilot in Office from $99.99 a year. |
| 7 | Perplexity | 65 | You will pay for Pro ($17 a month, annual) for research, images, video and files. |
| 8 | Meta AI | 64 | You want a free assistant that creates images and connects to your calendar and email. |
| 9 | DeepSeek | 39 | You want a free chat app with search and a reasoning mode. |
Here are two slides from this article's worked example, as NOXI built them, unedited.
How we ranked: seven printed rules, official pages only
Each assistant is rated 1 to 5 on seven criteria; points = rating ÷ 5 × weight, and the weights total 100. Every rating comes from an official vendor page opened on 22 October 2026, linked in the assistant's card.
| Code | Criterion (weight) | The rule, identical for every assistant |
|---|---|---|
| A1 | Free plan (20) | Count five capabilities documented for the free plan: web search, file uploads, image generation, file or document creation (a .docx, .pptx, .xlsx or PDF, or export to Docs or Slides; an in-chat canvas without a documented file export does not count), and connections to other apps. 5 = all five; 4 = four; 3 = three; 2 = two; 1 = none or one. |
| A2 | Entry price (10) | The cheapest paid tier per month; with annual billing, the annual price ÷ 12. 5 = $10 or less, or no paid tier needed; 4 = $10.01–15; 3 = $15.01–20; 2 = $20.01–30; 1 = over $30. |
| A3 | Long documents (10) | The largest context window the vendor publishes for its plan nearest $20 a month, on a consumer page that opens without logging in (API model pages do not count). 5 = 1M tokens or more; 3 = 128K to under 1M; 1 = under 128K; 2 = not published. |
| A4 | Tools on the ~$20 plan (20) | Count six tools documented for the plan nearest $20 a month on the monthly list price: web search, deep research, image generation, voice conversation (a spoken back-and-forth, not dictation), file creation, video generation. 5 = five or six; 4 = four; 3 = three; 2 = two; 1 = none or one. |
| A5 | Add-ons for individuals (15) | 5 = custom MCP connectors on free and paid personal plans; 4 = custom MCP on paid personal plans only, or with age or region limits; 3 = an app or plugin directory, with custom MCP only in business plans; 2 = built-in links to a few named services; 1 = none documented. |
| A6 | Training control (15) | 5 = the vendor states consumer chats are not used to train its models; 4 = an opt-out setting plus a chat mode excluded from training; 3 = an opt-out setting only; 2 = opt-out only by request or form; 1 = no opt-out stated. |
| A7 | Where it works (10) | Count six surfaces: web, desktop app, mobile app, a browser extension or the vendor's own browser, an office-suite add-in or integration, the vendor's own coding agent. 5 = six; 4 = five; 3 = four; 2 = three; 1 = two or fewer. |
Two clarifications apply to every vendor. A capability counts for a plan when a page lists it for that plan, or documents it with no plan limit; a plan-specific statement beats a general one. A desktop app counts when it runs the assistant, including agent apps such as Grok Bot, DeepSeek Harness or Perplexity for Mac.
Why the long-document rule gives a 2 to seven assistants
Seven vendors publish no consumer-plan context window on an open page, so all seven get the same 2; OpenAI's Go article says only "a larger context window". xAI, DeepSeek, OpenAI and Mistral publish API context windows, which describe developer access, not the chat plan. Google and Anthropic publish consumer figures.
AI chatbot comparison table: every cell and its arithmetic
Every rating, then the points. A 2 in long documents means "not published on an open page", not "short".
| Assistant | A1 Free plan (20) | A2 Entry price (10) | A3 Long docs (10) | A4 Tools (20) | A5 Add-ons (15) | A6 Training (15) | A7 Where (10) | Total |
|---|---|---|---|---|---|---|---|---|
| Google Gemini | 5 | 5 | 5 | 5 | 4 | 4 | 5 | 94 |
| ChatGPT | 5 | 5 | 2 | 5 | 4 | 4 | 5 | 88 |
| Claude | 4 | 3 | 5 | 4 | 5 | 4 | 5 | 85 |
| Grok | 4 | 5 | 2 | 5 | 5 | 4 | 3 | 83 |
| Mistral Vibe | 5 | 4 | 2 | 4 | 5 | 4 | 2 | 79 |
| Microsoft Copilot | 3 | 5 | 2 | 5 | 2 | 5 | 5 | 77 |
| Perplexity | 2 | 3 | 2 | 5 | 4 | 3 | 3 | 65 |
| Meta AI | 4 | 5 | 2 | 5 | 2 | 2 | 1 | 64 |
| DeepSeek | 2 | 5 | 2 | 1 | 1 | 2 | 2 | 39 |
| # | Assistant | Points: A1 + A2 + A3 + A4 + A5 + A6 + A7 | Total |
|---|---|---|---|
| 1 | Google Gemini | 20 + 10 + 10 + 20 + 12 + 12 + 10 | 94 |
| 2 | ChatGPT | 20 + 10 + 4 + 20 + 12 + 12 + 10 | 88 |
| 3 | Claude | 16 + 6 + 10 + 16 + 15 + 12 + 10 | 85 |
| 4 | Grok | 16 + 10 + 4 + 20 + 15 + 12 + 6 | 83 |
| 5 | Mistral Vibe | 20 + 8 + 4 + 16 + 15 + 12 + 4 | 79 |
| 6 | Microsoft Copilot | 12 + 10 + 4 + 20 + 6 + 15 + 10 | 77 |
| 7 | Perplexity | 8 + 6 + 4 + 20 + 12 + 9 + 6 | 65 |
| 8 | Meta AI | 16 + 10 + 4 + 20 + 6 + 6 + 2 | 64 |
| 9 | DeepSeek | 8 + 10 + 4 + 4 + 3 + 6 + 4 | 39 |
No two assistants tie; a tie would be listed alphabetically. Points = rating ÷ 5 × weight; documented capability on official pages checked 22 October 2026.
How sensitive the order is
Five cells rest on a page behind a login or one that reads two ways:
- ChatGPT, long documents: the context window for each plan sits behind a login. If OpenAI publishes 1M tokens or more on an open page, ChatGPT scores 94; if it publishes 128K to under 1M, ChatGPT scores 90.
- ChatGPT, free plan: file creation on Free rests on OpenAI's statement that ChatGPT Work is included in the Free plan; the Free plan card says Work runs in the desktop app, subject to rollout. Without that capability ChatGPT scores 84 and Claude leads it 85 to 84.
- Grok: the free-plan cell rests on xAI's "free to try" statement (4 points fewer for each of web search and image generation not counted, 8 if neither is), and the surfaces cell counts the Grok Bot desktop app (2 points fewer if not). Grok's range is 73 to 83.
- Perplexity, add-ons: the custom remote-connector article is filed under Enterprise, while the personal custom-plugin article supports a 4. Read as Enterprise-only, Perplexity scores 62. No other position changes.
Best AI chatbot for each need, from the scored criteria
The best AI chatbot for a given need is the assistant that scores 5 in that need's column.
| Need | Column | Who scores 5 | The documented fact |
|---|---|---|---|
| Best free plan | A1 | Gemini, ChatGPT, Mistral Vibe | Web search, uploads, images, file creation and app connections all documented for free use. |
| Entry price of $10 or less | A2 | Gemini, ChatGPT, Grok, Copilot, Meta AI, DeepSeek | Google AI Plus $4.99; ChatGPT Go $8; X Subscription Plus $8; Microsoft 365 Personal $8.33 on annual billing; Meta AI and DeepSeek free. |
| Long documents | A3 | Gemini, Claude | 1M tokens on Google AI Pro; "Up to 1M, varies by model" on Claude Pro. |
| Most tools on a ~$20 plan | A4 | Gemini, ChatGPT, Grok, Copilot, Perplexity, Meta AI | Five or six of: web search, deep research, images, voice, file creation, video. |
| Your own connectors | A5 | Claude, Grok, Mistral Vibe | Custom MCP connectors on free and paid personal plans. |
| Privacy by default | A6 | Microsoft Copilot | The vendor states that prompts, responses and files in the new Copilot app are not used to train foundation models. |
| Everywhere you work | A7 | Gemini, ChatGPT, Claude, Copilot | Web, desktop, mobile, browser, office suite and a coding agent. |
ChatGPT vs Claude vs Gemini: the three-way comparison
Gemini scores 94, ChatGPT 88 and Claude 85. Gemini has no cell below ChatGPT and trails Claude only on add-ons; ChatGPT and Claude reach their totals from opposite strengths.
| Fact (22 Oct 2026) | ChatGPT | Claude | Google Gemini |
|---|---|---|---|
| Cheapest paid plan | Go, $8 a month, monthly only | Pro, $17 a month billed annually ($200 up front) or $20 monthly | Google AI Plus, $4.99 a month |
| ~$20 plan | Plus, $20 a month | Pro, $20 monthly | Google AI Pro, $19.99 a month |
| Free plan (of 5 capabilities) | 5: search, uploads, images, file creation through Work, plugins | 4: search, uploads, file creation, connectors; no image generation documented | 5: Deep Research, uploads, images, slide generation, connected apps |
| Context window, ~$20 plan | Not published on an open page | Up to 1M tokens, varies by model | 1M tokens (32K without a plan, 128K on AI Plus) |
| Tools on ~$20 plan (of 6) | 5: video generation not documented | 4: image and video generation not documented | 6 |
| Add-ons | Plugin directory on all plans; custom MCP with read/fetch permissions on Pro; full MCP on Business and Enterprise/Edu | Custom connectors on every plan; Free limited to one | Custom MCP apps for personal accounts, age 18+ and in the US |
| Training control | Opt-out toggle plus Temporary chats | Opt-out plus Incognito chats | Keep Activity off plus Temporary chats |
| Where it works | 6 of 6 | 6 of 6 | 6 of 6 |
| Score | 88 | 85 | 94 |
Where ChatGPT and Claude differ. ChatGPT gains 4 points each on the free plan (A1), the entry price (A2) and tools (A4), 12 in all. Claude gains 6 on the published long-document limit (A3) and 3 on add-ons (A5), 9 in all, so ChatGPT leads 88 to 85. They tie on training control and surfaces. The full head-to-head, with plans, usage and privacy settings, is in ChatGPT vs Claude.
Where ChatGPT and Gemini differ. Gemini leads 94 to 88, only on long documents: 6 points for a published 1M-token window against none published. ChatGPT has no cell above Gemini; the two tie on the other six criteria, add-ons included. The detail is in ChatGPT vs Gemini.
ChatGPT vs Claude vs Gemini vs Grok, Perplexity and Copilot
Adding three more changes nothing at the top: Grok scores 83, Microsoft Copilot 77 and Perplexity 65. Grok beats ChatGPT on add-ons, Copilot has the only 5 for training control, and Perplexity documents five tools on Pro but a narrower free plan.
Claude vs Gemini: four documented differences
Gemini scores 94 and Claude 85; the 9-point gap comes from three cells where Gemini leads by 4 points each and one where Claude leads by 3. They tie on long documents, training control and surfaces.
- Entry price (Gemini +4). Google AI Plus costs $4.99 a month; Claude's cheapest paid plan is Pro at $200 a year, $16.67 a month.
- Free plan (Gemini +4). Google's help centre ticks image generation for users without a plan; no Claude page we opened documents image generation.
- Tools on the ~$20 plan (Gemini +4). Google AI Pro documents all six tools, video generation included; Claude Pro documents four.
- Add-ons (Claude +3). Claude offers custom connectors on every plan, Free included with a limit of one; Gemini's custom MCP apps require you to be 18 or over and in the US.
One difference outside the scores matters for long files: Claude lists "Up to 1M" tokens on every plan, Free included; Gemini without a plan has 32K.
The best free AI chatbot: what each free plan documents
The best free AI chatbots on documented capability are Google Gemini, ChatGPT and Mistral Vibe, which each document all five free-plan capabilities we count. Claude, Grok and Meta AI document four; Microsoft Copilot three; Perplexity and DeepSeek two.
| Assistant | Documented on the free plan | Not documented for free use | A1 |
|---|---|---|---|
| Google Gemini | Deep Research (web search), uploads, images, slide generation, connected apps | — | 5 |
| ChatGPT | Web search, uploads, image creation, plugins, ChatGPT Work file creation (desktop, subject to rollout) | — | 5 |
| Mistral Vibe | Web searches, uploads, basic images, 100+ connectors, Canvas export to PowerPoint and Excel | — | 5 |
| Claude | Web search, uploads, file creation, connected apps and tools | Image generation | 4 |
| Grok | Real-time web search, image generation, file and PDF analysis, connectors | File export | 4 |
| Meta AI | Image creation, photo editing from uploads, web research, calendar and email connections | File export from Canvas | 4 |
| Microsoft Copilot | Chat with image generation (limited usage), uploads, web search | Connectors (Microsoft 365 plans only), file creation | 3 |
| Perplexity | Basic searches with no practical cap, limited uploads | Image generation, file creation (Pro), connectors | 2 |
| DeepSeek | Search and DeepThink modes, uploaded files | Images, file creation, connectors | 2 |
Free plans carry usage limits described in words ("limited usage"), not counts. For paid plans compared on yearly cost, see our best AI subscription ranking.
The 9 assistants: plans, strengths and every rating's source
Prices are US list prices on 22 October 2026; local prices differ.
1. Google Gemini 94 / 100
Plans. No plan $0, Google AI Plus $4.99 a month, AI Pro $19.99, AI Ultra from $99.99 (subscriptions page).
Free plan (5). The help-centre feature table ticks Canvas, Connected apps, Slide generation, Deep Research (multi-step web search) and Image generation without a plan; uploads work for any signed-in user. Entry price (5). AI Plus, $4.99. Long documents (5). 1M tokens on AI Pro; 128K on AI Plus, 32K without a plan. Tools (5). AI Pro: Deep Research, images, a limited video-generation trial, slides, Gemini Live voice and search. Add-ons (4). Custom MCP apps on personal accounts for users 18 or over in the US with Keep Activity on. Training (4). With Keep Activity off, chats "won't be used to train our AI models"; Temporary chats are excluded too (Privacy Hub, updated 19 October 2026). Where (5). Web, Mac and Windows apps, mobile, Chrome, Gmail and Docs, Jules and Antigravity.
Best when: you want the most documented features per dollar and already use Google's apps.
2. ChatGPT 88 / 100
Plans. Free, Go $8 a month, Plus $20, Pro from $100 ($100, $200 or $500), all monthly only (pricing docs; Go).
Free plan (5). The Free Tier FAQ lists web search, uploads and image creation; the plugin directory is on every plan; ChatGPT Work, included in Free, creates documents, spreadsheets and presentations. Entry price (5). Go, $8. Long documents (2). Not published on an open page. Tools (5). Plus: voice, images, uploads, deep research, web search and Work; video not documented. Add-ons (4). Pro users can connect custom MCP servers with read/fetch permissions; full MCP is for Business and Enterprise/Edu. Training (4). "Improve the model for everyone" toggle plus Temporary chats (Data controls). Where (5). Web, desktop, mobile, Chrome browser control, ChatGPT for Excel, Codex (OpenAI help).
Best when: you want a full free plan, five tools on Plus and one assistant on six surfaces.
3. Claude 85 / 100
Plans. Free, Pro $17 a month billed annually ($200 up front) or $20 monthly, Max from $100 (Claude pricing).
Free plan (4). "Search the web, create files, and run code" and "Connect your apps and tools"; the upload article sets no plan limit; image generation is not documented. Entry price (3). $200 ÷ 12 = $16.67. Long documents (5). "Up to 1M, varies by model" (context window). Tools (4). Web search, Research, Voice mode, file creation. Add-ons (5). Custom connectors on every plan; one on Free. Training (4). Opt-out on Free, Pro and Max, plus Incognito chats. Where (5). Web, desktop, mobile, Claude in Chrome, Claude for Microsoft 365, Claude Code.
Best when: you work with long documents and want to connect your own tools, even on the free plan.
4. Grok 83 / 100
Plans. The X Subscription: Free, Plus $8 a month or $80 a year, Premium $30 or $300 a year, Super $100, Ultra $200 (xAI pricing).
Free plan (4). The product page lists real-time web search, image generation and file and PDF analysis, "free to try"; connectors are open to all users; no file export is documented. Entry price (5). Plus, $8. Long documents (2). Not published. Tools (5). Premium, the tier nearest $20: images and video, voice, web search, and Grok Bot documents, spreadsheets and decks. Add-ons (5). Custom MCP via New Connector, then Custom. Training (4). "Improve the Model" setting; Private Chat is not used for training (legal FAQ). Where (3). Web, mobile, Grok Bot desktop, Grok Build.
Best when: you want custom connectors on every plan and a paid tier from $8.
5. Mistral Vibe 79 / 100
Plans. Free, Pro $14.99 a month, Team $24.99 per user (Mistral pricing). Mistral now calls its assistant Vibe, formerly Le Chat.
Free plan (5). Web searches, document upload, basic image generation, 100+ connectors, and Canvas export of a deck as PowerPoint or a table as CSV or Excel. Entry price (4). Pro, $14.99. Long documents (2). Not published. Tools (4). Web search, Deep Research, images, spreadsheets and slides; voice mode is transcription, and no video is listed. Add-ons (5). Custom MCP on Free, Pro and Student. Training (4). Opt-out plus Temporary Chat (Vibe docs). Where (2). Web, mobile, Vibe Code.
Best when: you want a full free plan with PowerPoint and Excel export and your own MCP connectors.
6. Microsoft Copilot 77 / 100
Plans. Free Microsoft account; Microsoft 365 Personal $99.99 a year ($9.99 monthly), Family $129.99, Premium $199.99 ($19.99 monthly) (compare page).
Free plan (3). Chat and image generation with "limited usage", uploads and web search; connectors are for Microsoft 365 plans. Entry price (5). $99.99 ÷ 12 = $8.33. Long documents (2). Not published. Tools (5). Premium: image creation, advanced research, Copilot in Word and PowerPoint, Copilot Voice, web search. Add-ons (2). OneDrive, Outlook.com, Google Drive, Gmail. Training (5). Prompts, responses and files in the new app "aren't used to train foundation models" (activity history, app updated 18 August 2026; same for the home apps). Where (5). Web, desktop, mobile, Edge, Office apps, GitHub Copilot.
Best when: you want the only 5 for training control in this ranking and Copilot inside Office.
7. Perplexity 65 / 100
Plans. Free, Pro $17 a month billed annually, Max $167 a month billed annually (Perplexity Pro).
Free plan (2). Basic searches with no practical cap and limited uploads; images and file creation are Pro (help). Entry price (3). $17. Long documents (2). Not published. Tools (5). Pro Search, deep research, image and video generation, files and apps (help). Add-ons (4). Custom plugins with remote MCP servers on plans with Perplexity Computer, Pro included. Training (3). Opt-out in Account Settings. Where (3). Web, Mac, iPhone, Comet (page).
Best when: you will pay for Pro and want research, images, video and files together.
8. Meta AI 64 / 100
Plans. Free for everyday use; Meta One Core $7.99 a month and Premium $19.99, where available (Meta, 15 September 2026).
Free plan (4). Image creation, photo editing from uploads, web research and calendar and email connections (meta.ai; help; help). Entry price (5). Free; Core $7.99. Long documents (2). Not published. Tools (5). Premium: deep research, web search, images and video, voice (help). Add-ons (2). Built-in connections. Training (2). An EU objection form (Meta, updated 27 March 2026). Where (1). Web and app.
Best when: you want a free assistant that makes images and connects to your calendar and email.
9. DeepSeek 39 / 100
Plans. Free: "Chat with leading AI models for free", with no paid tier (download page).
Free plan (2). DeepThink and Search modes on the official site; uploaded files per the privacy policy. Entry price (5). No paid tier needed. Long documents (2). API figures only. Tools (1). Web search. Add-ons (1). None documented. Training (2). Opt-out by email (privacy policy, 10 February 2026). Where (2). Web, mobile, DeepSeek Harness desktop.
Best when: you want a free chat app with search and a reasoning mode and no subscription.
ChatGPT alternatives: when to switch, and to which
The strongest documented ChatGPT alternative is Google Gemini (94 against 88); Claude scores 85, three points behind, with different strengths. Switch for a cell where the alternative scores higher, and check what you give up.
| Alternative | Score | Scores above ChatGPT on | Scores below ChatGPT on |
|---|---|---|---|
| Google Gemini | 94 | Long documents (+6) | None |
| Claude | 85 | Long documents (+6), add-ons (+3) | Free plan (−4), entry price (−4), tools (−4) |
| Grok | 83 | Add-ons (+3) | Free plan (−4), where it works (−4) |
| Mistral Vibe | 79 | Add-ons (+3) | Entry price (−2), tools (−4), where it works (−6) |
| Microsoft Copilot | 77 | Training control (+3) | Free plan (−8), add-ons (−6) |
For a free alternative, Gemini and Mistral Vibe document all five free-plan capabilities. Mistral vs ChatGPT (Le Chat vs ChatGPT) is 79 against 88; Meta AI vs ChatGPT is 64 against 88, level on price and tools.
ChatGPT vs Copilot
ChatGPT scores 88 and Microsoft Copilot 77. ChatGPT leads on the free plan by 8 points (five capabilities against three) and on add-ons by 6; Copilot leads on training control by 3, as the only assistant whose vendor states that consumer chats are not used to train its models. For PowerPoint work, see our Copilot vs Gemini vs ChatGPT for presentations test.
ChatGPT vs Perplexity
ChatGPT scores 88 and Perplexity 65. ChatGPT leads on the free plan (+12), entry price (+4), training control (+3) and surfaces (+4); they tie on add-ons, since both document custom MCP servers on Pro. On the ~$20 plans both document five tools, and Perplexity's include video generation.
DeepSeek vs ChatGPT
DeepSeek scores 39 and ChatGPT 88. Both cost nothing to start; ChatGPT leads on the free plan (+12), tools (+16), add-ons (+9), training control (+6) and surfaces (+6). DeepSeek's privacy policy says personal data is collected, processed and stored in the People's Republic of China, which some readers will weigh before uploading files.
Grok vs ChatGPT
ChatGPT scores 88 and Grok 83. ChatGPT leads on the free plan (+4) and surfaces (+4); Grok leads on add-ons (+3), since custom connectors are open to every Grok user. Both start at $8 a month and document five tools on the tier nearest $20, which for Grok is Premium at $30.
Which AI is best for writing, research, studying and work
For long writing, pick an assistant that publishes a 1M-token window (Gemini on AI Pro, Claude); for research, seven name a deep-research tool on the ~$20 plan; for studying, the three full free plans lead; for work, Copilot and Claude pair office integrations with training controls. No page measures writing quality, so these picks rest on documented tools.
| Job | What to look for | Who documents it | Column |
|---|---|---|---|
| Writing long drafts | A published long-document limit and file creation | Gemini (1M on AI Pro), Claude (up to 1M); file creation on the ~$20 plan for seven of nine | A3, A4 |
| Research | Deep research on the ~$20 plan | Gemini, ChatGPT, Claude, Mistral Vibe, Copilot, Perplexity, Meta AI; Grok describes cited multi-agent answers | A4 |
| Studying | A full free plan, then a low entry price | Gemini, ChatGPT, Mistral Vibe (all five free capabilities); Gemini $4.99, ChatGPT Go $8 | A1, A2 |
| Work | Office-suite integration and training control | Copilot (Word, Excel, PowerPoint; no-training statement), Claude (Microsoft 365; opt-out and Incognito), ChatGPT (Excel), Gemini (Gmail, Docs) | A6, A7 |
| Connecting your tools | Custom MCP connectors on personal plans | Claude, Grok, Mistral Vibe (free and paid); Gemini (18+, US); ChatGPT and Perplexity on Pro | A5 |
Business plans, such as ChatGPT Business at $20 per user a month billed annually, are outside this consumer ranking. Before you paste client material into any assistant, read our checklist for confidential data in AI tools.
Most popular AI chatbot: who uses which, Pew 2026 and vendor figures
ChatGPT is the most-used chatbot in the United States: 44% of US adults say they ever use it, against 24% for Gemini and 17% for Microsoft Copilot (Pew Research Center, survey of 5,119 adults, 17–23 February 2026, published 17 June 2026) [1]. About half of US adults (49%) now use AI chatbots, and 24% use them daily or more often.
| Chatbot | US adults who ever use it (Pew 2026) | Official user figure from the vendor |
|---|---|---|
| ChatGPT | 44% | More than 1.2 billion people each week (OpenAI, 21 October 2026) [2] |
| Gemini | 24% | More than 1 billion people every month (Google, 11 August 2026) [3] |
| Copilot | 17% | No figure on the pages opened |
| Meta AI | 14% | No figure on the pages opened |
| Grok | 8% | No figure on the pages opened |
| Claude | 6% | No figure on the pages opened |
Two cautions. OpenAI counts weekly users and Google monthly users, so the two figures should not be set against each other. And Pew's 2026 question asked about several chatbots at once, a change in wording from earlier years, so do not compare 44% with older ChatGPT figures. Our guide to presenting survey results shows how to put figures like these on a slide with their base and dates.
How to re-weight the ranking for your own needs
To re-weight the ranking, move points between criteria so the weights still total 100, then recompute each assistant as rating ÷ 5 × new weight. The ratings stay the same; only your priorities change the order.
| Assistant | Points with Sofia's weights (A1 10, A6 25) | New total | Change |
|---|---|---|---|
| Google Gemini | 10 + 10 + 10 + 20 + 12 + 20 + 10 | 92 | −2 |
| ChatGPT | 10 + 10 + 4 + 20 + 12 + 20 + 10 | 86 | −2 |
| Claude | 8 + 6 + 10 + 16 + 15 + 20 + 10 | 85 | 0 |
| Grok | 8 + 10 + 4 + 20 + 15 + 20 + 6 | 83 | 0 |
| Microsoft Copilot | 6 + 10 + 4 + 20 + 6 + 25 + 10 | 81 | +4 |
| Mistral Vibe | 10 + 8 + 4 + 16 + 15 + 20 + 4 | 77 | −2 |
| Perplexity | 4 + 6 + 4 + 20 + 12 + 15 + 6 | 67 | +2 |
| Meta AI | 8 + 10 + 4 + 20 + 6 + 10 + 2 | 60 | −4 |
| DeepSeek | 4 + 10 + 4 + 4 + 3 + 10 + 4 | 39 | 0 |
Gemini still leads at 92 and ChatGPT stays second at 86, but its lead over Claude (85) shrinks from 3 points to 1, and Microsoft Copilot rises above Mistral Vibe to 81 because of its 5 for training control. Copilot's row in full: 3 ÷ 5 × 10 = 6, then 10, 4, 20, 6, 5 ÷ 5 × 25 = 25 and 10, for 81. A 4 still means Sofia must switch the setting off herself.
What this ranking does not score, and why
This ranking does not score answer quality, accuracy, writing style, speed or usage limits, because no vendor documents them in a form that can be checked alike for nine products on one day. Quality benchmarks are run by third parties on models, not on the consumer apps. Every assistant can be wrong: check any figure against its source before you rely on it.
What is the best AI model in 2026?
The best AI model changes month to month, and vendors rename and replace models faster than any ranking can follow, so this page scores the assistant you use rather than the model inside it. Each assistant's model list is on its own plan page; the plan you choose decides which models you get.
If your goal is a presentation: which AI is best for PowerPoint
For PowerPoint and slides, use a ranking built for presentations: our AI presentation tools ranking scores 20 tools on export, editing and sources, and the best AI presentation makers page reports our June 2026 hands-on test. Several assistants here document slide output: Gemini (slide generation), Claude (Slides on Free), Mistral Vibe (Canvas to PowerPoint), ChatGPT Work and Copilot in PowerPoint.
Our pick for turning a chat into a designed deck is NOXI, a presentation maker rather than a chatbot. NOXI is free: no card and no watermark, with a monthly allowance of 1,000 credits per account, and a full deck uses about 25. It builds the full deck from a topic, a brief or an outline, lets you change one slide or the whole deck in plain language, and downloads an editable PowerPoint (.pptx) with real text boxes and embedded fonts, or a PDF. You can also connect NOXI to your assistant with the address on noxi.ai/connect and ask for the deck in the chat; the flow is below.
Key terms
Build it: connect NOXI to Claude and turn a comparison into a deck
A shortlist for your team becomes a deck in one chat: add the NOXI connector in Claude, ask for the presentation, then open it in NOXI to edit, present or export. Type these prompts once the connector is added.
Using the NOXI connector, create a [9]-slide presentation for [our IT team] comparing [three] AI assistants with these scores: [paste your table]. Show the verdict first, a side-by-side table, the score arithmetic, a privacy rule for staff and a pilot plan with dates.
Research [how US adults use AI chatbots in 2026] and turn it into an [8]-slide NOXI deck for [our leadership team], with the source and date of every figure on the slide where it appears.
In my latest NOXI deck, rewrite every slide title as a full sentence that states the finding, and keep only the decision, the pilot dates and the review date on the last slide.
Step 1 — Create a free NOXI account
Open noxi.ai and create a free account; no card is needed.
Step 2 — Copy the server address from noxi.ai/connect
The page noxi.ai/connect gives one server address for every assistant it lists, mcp.noxi.ai/mcp. You sign in to NOXI in your browser, so there is no API key to copy.
Step 3 — Add a custom connector in Claude
In Claude, go to Customize, then Connectors, click "+ Add" and choose "Add custom connector". Claude's help centre says custom connectors work on Free, Pro, Max, Team and Enterprise plans, and Free users are limited to one.
Step 4 — Name the connector NOXI and sign in
Enter NOXI as the name, paste the server address as the remote MCP server URL and click Continue. Leave the advanced settings empty (no client ID is needed), click Add, then sign in to NOXI and approve access. NOXI only works with presentations in the account you connected.
Step 5 — Ask for the deck in the chat
Paste Prompt 1 with your own assistants and scores, and name NOXI in the request. The connector creates the designed presentation, and the PDF and PowerPoint files arrive in the chat.
Step 6 — Open it in NOXI to edit, present or export
Open the deck in NOXI to change one slide or the whole deck in plain language, present full-screen or share a link, and download an editable PowerPoint (.pptx) with real text boxes and embedded fonts, or a PDF.
ChatGPT and other MCP clients are also listed on noxi.ai/connect. You can disconnect at any time in your assistant's connector settings.
Frequently asked questions
- What is the best AI chatbot in 2026?
- On seven documented criteria checked on 22 October 2026, Google Gemini scores highest with 94 of 100, followed by ChatGPT at 88, Claude at 85 and Grok at 83. The score counts free-plan features, price, long-document limits, tools, add-ons, training controls and surfaces. It does not measure answer quality, which no vendor documents.
- Which AI is the best right now?
- Right now, Google Gemini documents the most of what this ranking counts: all five free-plan capabilities, a $4.99 entry plan, a 1M-token window on AI Pro and six tools at $19.99. If one feature matters most, pick by that column instead: Claude for custom connectors, Microsoft Copilot for its stated no-training rule.
- What is the best free AI chatbot?
- Google Gemini, ChatGPT and Mistral Vibe are the best free AI chatbots on documented capability: each lists web search, file uploads, image generation, file creation and app connections for free use. Claude, Grok and Meta AI document four of the five. Free plans still carry usage limits that vendors describe in words, not counts.
- Is ChatGPT still the most popular chatbot?
- Yes. In Pew Research Center's February 2026 survey, 44% of US adults said they ever use ChatGPT, against 24% for Gemini and 17% for Copilot. OpenAI said on 21 October 2026 that more than 1.2 billion people use ChatGPT each week; Google reports more than 1 billion monthly Gemini app users.
- ChatGPT vs Claude vs Gemini: which should I use?
- Use Gemini (94) for the broadest documented plan at the lowest price. Use ChatGPT (88) for a full free plan, five tools on Plus and six surfaces. Use Claude (85) for a published window of up to 1M tokens and custom connectors on every plan. ChatGPT leads by 3; without free-plan file creation, Claude would lead 85 to 84.
- Is Claude better than Gemini?
- On this rubric, Gemini scores 94 and Claude 85. Gemini leads on entry price ($4.99 against $16.67 a month), the free plan and tools, including image and video generation. Claude leads on add-ons, with custom connectors on every plan including Free. Both publish a 1M-token window and score the same on privacy and surfaces.
- What is a good ChatGPT alternative?
- Google Gemini is the highest-scoring ChatGPT alternative at 94, ahead on long documents with no cell below ChatGPT. Claude scores 85, three points behind, and leads on long documents and connectors. For a free alternative, Gemini and Mistral Vibe document the same five free-plan capabilities as ChatGPT.
- Is DeepSeek safe to use?
- DeepSeek's privacy policy, last updated 10 February 2026, says it collects, processes and stores personal data in the People's Republic of China and lists uploaded files as user input. Opting out of training is done by email; no setting is named. Weigh those facts before sharing documents, and avoid sharing confidential material.
- Which AI is best for research?
- For research, Gemini, ChatGPT, Claude, Mistral Vibe, Microsoft Copilot, Perplexity and Meta AI name a deep-research tool on their plan nearest $20 a month, and Grok describes multi-agent answers with citations. Gemini and Claude also publish a 1M-token window for long sources. Whichever you choose, open the cited pages before you rely on a figure.
- Which AI chatbot can make a PowerPoint?
- Several document slide output: Mistral Vibe's Canvas exports a deck as PowerPoint, ChatGPT Work creates presentations, Claude's Free plan lists Slides, Gemini's free tier ticks slide generation, and Copilot works in PowerPoint on Microsoft 365 Premium. For designed decks, see our presentation tools ranking, which scores 20 tools on export and editing.
Sources & methodology
Every vendor page below was opened on 22 October 2026, logged out; each rating's page is also linked in the assistant's card. Sofia Reyes is a fictional example.
- [1] Chatbot use among US adults (49%, 24% daily; ChatGPT 44%, Gemini 24%, Copilot 17%, Meta AI 14%, Grok 8%, Claude 6%) — Pew Research Center, Americans and AI 2026: Chatbots, Smart Devices and Views on Impact (17 June 2026; survey 17–23 February 2026, n = 5,119); checked 22 October 2026.
- [2] ChatGPT weekly users — OpenAI, ChatGPT Ads expands to 26 new countries (21 October 2026); checked 22 October 2026.
- [3] Gemini app monthly users — Google, The Gemini app has 1 billion monthly users (11 August 2026); checked 22 October 2026.
- [4] Google Gemini — Google AI subscriptions (US); Gemini Help, features and limits by plan; Gemini Help, upload files; Gemini Help, custom MCP apps; Gemini Apps Privacy Hub (updated 19 October 2026); Gemini Help, Gemini app on Mac and Windows; checked 22 October 2026.
- [5] ChatGPT — OpenAI, ChatGPT pricing docs; ChatGPT Free Tier FAQ; What is ChatGPT Go; What is ChatGPT Plus; Plugins in ChatGPT; ChatGPT Work; Connecting custom MCP servers; Data controls in ChatGPT; OpenAI Help Center, article 20001063; chatgpt.com/pricing redirected to a sign-in page; checked 22 October 2026.
- [6] Claude — Claude pricing; Claude Help, uploading files; Claude Help, context window; Claude Help, custom connectors using remote MCP; Claude Help, incognito chats; checked 22 October 2026.
- [7] Grok (xAI) — xAI, Grok; xAI docs, The X Subscription pricing; Grok plans; xAI docs, connectors; xAI docs, Grok FAQ; xAI docs, Grok Bot; xAI docs, Grok Bot files and results; xAI docs, Grok Build; xAI legal FAQ; checked 22 October 2026.
- [8] Mistral Vibe — Mistral AI pricing; Mistral docs, Vibe; Mistral docs, files and Canvas; Mistral docs, voice mode; Mistral docs, MCP connectors; Mistral Help, training opt-out; checked 22 October 2026.
- [9] Microsoft Copilot — Microsoft 365, compare all plans; Microsoft Copilot features for individuals; Microsoft Support, file upload in Copilot; Microsoft Support, Privacy FAQ for Microsoft Copilot; Microsoft Support, connecting Copilot to other services; Microsoft Support, Copilot Voice; Microsoft Support, Copilot activity history and training; Microsoft Support, Copilot in Microsoft 365 apps for home: your data and privacy; Microsoft Edge, Copilot; checked 22 October 2026.
- [10] Perplexity — Perplexity Pro; Perplexity Help, plans; Perplexity Help, Pro features; Perplexity Help, create or upload a custom plugin; Perplexity, personal computer; checked 22 October 2026.
- [11] Meta AI — meta.ai; Meta Help, connecting apps; Meta Help, Meta AI features; Meta Help, Meta AI app; Meta, Introducing Meta One (15 September 2026); Meta, Making AI work harder for Europeans (14 April 2025, updated 27 March 2026); checked 22 October 2026.
- [12] DeepSeek — DeepSeek; DeepSeek download; DeepSeek Privacy Policy (last updated 10 February 2026); checked 22 October 2026.
- [13] NOXI connector address, sign-in and privacy lines — NOXI, Connect NOXI to your AI assistant; checked 22 October 2026.
Pick the assistant whose top-scoring column is the feature you need, then turn your comparison into a deck: NOXI is free, connects to your assistant from noxi.ai/connect, and downloads an editable PowerPoint.
