GeminiBest overall
Deep Research and Gemini Live free, AI Plus at $4.99 and AI Pro at $19.99 with 5 TB of storage; Gemini 3.8 Flash ranks 8th on LMArena. Keep Activity trains on chats by default.
AI assistants
An AI assistant is a chat app built on a large language model that answers questions, writes, analyses files and runs multi-step tasks. These AI assistants reviews compare eight of them, from free DeepSeek to Claude Pro at $20 a month, on model rankings, features, privacy settings and price, including what changes when you pay.
ClaudeHighest-rated AI models
Claude Opus 5.5 is 2nd on LMArena and 1st on the Artificial Analysis index, and Free runs Sonnet 5.5; no ads, but tighter limits and no storage bundle.
ChatGPTMost complete feature set
Work mode agents, Codex, deep research and voice on every platform, with plans from $8 to $500; its chat model trails Claude and Gemini, and Free and Go show ads.
Microsoft CopilotBest for Microsoft 365 users
Copilot inside Word, Excel, PowerPoint and Outlook from $9.99 a month with Microsoft 365 Personal, plus 1 TB of storage and separate training opt-outs.
PerplexityBest for cited research
Sources behind every answer and models from OpenAI, Anthropic, Google and Moonshot on Pro for $200 a year; trains on chats by default and faces publisher lawsuits.
Mistral Vibe (Le Chat)Best for EU data hosting
Mistral's assistant, renamed Vibe in May 2026, hosts data in Sweden and charges $14.99 for Pro; its models rank far below the leaders.
GrokBest for real-time X content
Live search of X posts, but SuperGrok costs $30 and Grok is under investigation by Ofcom, the European Commission and Ireland's DPC.
DeepSeekBest free open-weight models
Free V4 models that rank 39th on LMArena and can be self-hosted; the app stores data in China and has faced regulator action in several countries.
Compare side by side
Every score and price in one table. Select a column to sort. Prices are US list prices; hover a price for the full terms.
| Gemini | 8.6 | 8.8 | 9.0 | 6.8 | 9.4 | 9.3 | $4.99/mo | Yes | – |
|---|---|---|---|---|---|---|---|---|---|
| Claude | 8.5 | 9.6 | 8.8 | 7.4 | 8.2 | 8.4 | $17/mo | Yes | – |
| ChatGPT | 8.5 | 8.6 | 9.4 | 6.6 | 8.6 | 9.5 | $8/mo | Yes | – |
| Microsoft Copilot | 8.1 | 8.2 | 8.0 | 7.2 | 8.2 | 9.0 | $9.99/mo | Yes | – |
| Perplexity | 8.0 | 8.4 | 8.8 | 6.4 | 8.0 | 8.6 | $16.67/mo | Yes | – |
| Mistral Vibe (Le Chat) | 7.6 | 6.4 | 7.8 | 7.8 | 9.0 | 7.2 | $14.99/mo | Yes | – |
| Grok | 7.1 | 7.8 | 8.0 | 4.8 | 7.0 | 7.6 | $10/mo | Yes | – |
| DeepSeek | 6.2 | 7.6 | 5.4 | 2.8 | 9.0 | 6.0 | Free | Yes | – |
What separates one AI assistant from another?
It comes down to three things. The model on your plan, what happens to your chats and what else the subscription bundles. Every major assistant now searches the web, reads PDFs and drafts text, from the free DeepSeek app up to Perplexity Max at $200 a month. Those features decide nothing.
The model gap is real, though. Claude Opus 5.5 ranks 2nd on LMArena and has the highest score on the Artificial Analysis index, while Mistral Large 4 sits 115th. Don’t trust a vendor’s own benchmark chart over these public boards. Free plans usually run a smaller model than the one in the headlines. Gemini Free uses 3.5 Flash-Lite. ChatGPT Free and Go use GPT-6 Luna. When you compare plans, check the model name on the pricing page.
Which AI assistant fits you?
The default is Gemini. It ranks first overall. Its free plan includes Deep Research and Gemini Live, and AI Pro at $19.99 a month adds 5 TB of storage. Switch when one of these is true:
- If you want the strongest model for hard writing and reasoning, choose Claude. Pro costs $20 a month or $200 a year, with no ads. Its limits run out sooner than rivals’.
- If you want the most tools in one app, choose ChatGPT: agents in Work mode, Codex for code, the Atlas browser, deep research and voice. Plus costs $20 a month.
- If you live in Word, Excel and Outlook, Microsoft 365 Personal at $9.99 a month puts Copilot inside those apps and adds 1 TB of storage.
- If every answer needs a source, Perplexity cites the pages behind each reply, and Pro gives you models from several makers for $200 a year.
- If your data must stay in Europe, Mistral’s Vibe, formerly Le Chat, hosts data in Sweden. Pro costs $14.99 a month.
| Assistant | Free plan model | Main paid plan | Training on your chats |
|---|---|---|---|
| Gemini | 3.5 Flash-Lite | AI Plus $4.99/mo; AI Pro $19.99/mo | On while Keep Activity is on |
| ChatGPT | GPT-6 Luna (with ads) | Go $8/mo (ads); Plus $20/mo | On by default; opt-out in Data controls |
| Claude | Sonnet 5.5 | Pro $20/mo or $200/yr | Opt-out in Privacy settings |
| Microsoft Copilot | Copilot (free) | Microsoft 365 Personal $9.99/mo | Two separate opt-outs |
| Perplexity | Free tier | Pro $20/mo or $200/yr | On by default; opt-out in Preferences |
When should you not pay for an AI assistant?
Skip the upgrade if you ask a few questions a day. Gemini’s free plan already includes Deep Research, and Claude Free runs Sonnet 5.5. Don’t bother with a second assistant either. Two $20 plans mostly duplicate each other, so when you hit a real limit, pay for the one plan that removes it and cancel nothing you rely on for storage or Office. And don’t use DeepSeek’s free app for anything confidential: its privacy policy stores data in the People’s Republic of China.
What do most people miss before they subscribe?
Most people miss the settings, not the features.
- Your chats may train the model. Every assistant we rank trains on consumer chats by default or asks you to choose. For ChatGPT, turn off “Improve the model for everyone” under Data controls before you paste work data.
- Ads arrived on cheap plans. ChatGPT Free and Go have shown ads since February 2026; most rivals show none.
- The bundle changes the maths. Google AI Pro includes 5 TB of storage. Microsoft 365 Personal includes Office apps. Count what you already pay for storage and Office before you compare.
- Business plans work differently. Team plans from OpenAI, Anthropic, Google and Microsoft keep your data out of training by default. If you’ve got staff using personal accounts, that is the cheaper fix.
How should you read our AI assistants reviews?
Each score is a research score built from leaderboards, plan pages, privacy policies and regulator records. We don’t run our own prompt tests. In practice, read two lines: answer quality and privacy. Claude leads on quality at 9.6. Its total of 8.5 still trails Gemini’s 8.6, because Gemini scores 9.4 on price and Claude bundles no storage.
Low scores usually come from one weak line. DeepSeek’s privacy score is 2.8. Grok’s is 4.8, while Ofcom, the European Commission and Ireland’s DPC investigate it. The full ChatGPT review shows the method on one product. For code, see AI coding tools; for editing inside every app, see AI writing tools. Both sit in our AI tools section.
How we score AI assistants
Every product here is scored 0 to 10 on the same criteria, using published evidence. The overall score is the weighted average.
Answer quality & modelsWhich models each plan actually gives you (free and paid), and where those models place on independent leaderboards: the LMArena text ranking and the Artificial Analysis Intelligence Index, both checked in October 2026. Models announced but not yet in the app don't count.weight25%
Features & toolsWeb search with citations, deep research, file and data analysis, voice, agents and computer use, memory, projects, connectors and coding tools, and which plan each one needs. Image and video generation are not scored.weight20%
Privacy & data controlWhether consumer chats train the models by default and how to opt out, retention periods, human review, ads, where data is stored, protections on business plans, and regulator actions or security incidents.weight20%
Price & limitsFree plan, cheapest paid plan, the $20-tier value, team price per seat, bundled extras such as storage, and how clearly usage limits are published.weight20%
Apps & ecosystemWeb, desktop and mobile apps, browser and operating-system integration, office-suite and app connectors, admin tools for teams, and country availability.weight15%
Questions people ask
Which AI assistant is best for most people?
Gemini, on our evidence. It scores 8.6, the free plan includes Deep Research and Gemini Live, and AI Plus starts at $4.99 a month. Claude has the higher-rated model and ChatGPT the widest feature set, both at $20 a month. Turn off Keep Activity if you don't want Gemini to train on your chats.
Is a paid AI assistant worth $20 a month?
Only when you hit the free plan's limits. Paid plans unlock bigger models, higher usage caps and extras such as storage. If you've got a Google One or Microsoft 365 subscription already, check its AI tier before buying a separate plan, because Google AI Pro and Microsoft 365 bundle storage with the assistant.
Do AI chatbots use my conversations for training?
Usually yes on consumer plans unless you change a setting. ChatGPT, Claude, Gemini, Copilot and Perplexity all offer an opt-out or a choice. Business and team plans from OpenAI, Anthropic, Google and Microsoft exclude your data from training by default. Don't use a personal account with training switched on for client secrets.
Do AI assistants show ads?
Most don't. ChatGPT is the main exception: its Free and Go plans have shown ads chosen partly from your conversation since February 2026, and Go costs $8 a month. Claude carries no ads on any plan. If ads in a work tool bother you, skip the Go plan and compare Plus with Claude Pro, both $20 a month.
Is DeepSeek safe to use?
Treat the app with care. DeepSeek's privacy policy stores data in the People's Republic of China, and in January 2025 Wiz Research found an exposed DeepSeek database holding more than a million log lines, including chat history. The open-weight models are a different case: you can run them on your own servers, so prompts never reach DeepSeek.