Snapshot caveat: This file reflects prices and tiers as of June 2026. Subscription prices, tier names, and the model versions behind each app change frequently. The dollar amounts are more stable than the model-version labels. Re-verify on each provider's official pricing page before quoting.
Chat Apps and Subscriptions
Using AI
7 min read
Snapshot
In Short
The simplest way to use AI is a chat app: a website or phone app where you type and the company runs the model for you. ChatGPT, Claude, Gemini, Grok, Copilot, Meta AI, Perplexity, DeepSeek, and Kimi all offer this. Most have a free tier with daily limits and a paid tier around 20 US dollars a month that raises the limits and unlocks the best models and extra features. A few offer power-user tiers at 100 to 300 US dollars a month. For everyday use, a free tier or a single 20-dollar plan covers almost everyone.
01. What It Is
A chat app is a finished AI product you use through a browser or a phone app. You sign in, type a message, and get an answer. The large language model runs on the provider's servers, not your device, so any computer or phone can use the most powerful models.
This is the "subscription" path described in how-to-use-an-llm, and it is the one most people should choose.
Each app bundles a model (or several) with a polished interface and extra tools layered on top: web search, file and image uploads, image and video generation, voice conversation, and "deep research" modes that take longer but search more.
02. Why It Matters
The chat app is where the abstract idea of an LLM becomes something a normal person actually uses. The choice between them is usually small, the leading apps are broadly similar in everyday quality, so the practical questions are which interface you prefer, which extra features you want, which free-tier limits you keep hitting, and whether you already pay for a bundle (a Google or Microsoft subscription) that includes one. Understanding the free-versus-paid line saves money, since most people never need more than the free tier or a single paid plan.
03. How It Works: The Major Chat Apps
Tiers below are the consumer plans.
Each app also has a separate developer API (see how-to-use-an-llm) which is not covered here.
OpenAI ChatGPT
The best-known chat app, running OpenAI's GPT-5 family of models.
- Free: access to the default model with a usage cap, web search, and image creation. US users may see ads.
- Go (about 8 US dollars a month): a lower-cost paid tier with roughly ten times the free message allowance.
- Plus (about 20 US dollars a month): the mainstream paid tier. Higher limits, the latest models, deep research, image and video tools, and agent features.
- Pro (100 to 200 US dollars a month): the power-user tier, sold in two levels, 100 US dollars for five times the Plus limits and 200 US dollars for twenty times, with near-unlimited use, the largest context window, and the top reasoning model.
Anthropic Claude
Known for strong writing, careful instruction following, and document analysis. Runs the Claude Haiku, Sonnet, and Opus models.
- Free: chat, web search, file handling, and extended thinking, with daily limits, across web, phone, and desktop apps.
- Pro (20 US dollars a month, or about 17 a month billed yearly): more usage, Claude Code, projects, and research.
- Max (from 100 US dollars a month): five times or twenty times the Pro usage, higher output limits, and priority access at busy times.
Google Gemini
Google's chat app, tightly tied to its Google One storage subscriptions and the rest of Google's products. Runs the Gemini 3.x models.
- Free: the current Flash model plus limited access to the stronger Pro model, image generation, and a research mode, with 15 GB of Google storage.
- Google AI Plus (about 8 US dollars a month): roughly double the free limits and more storage.
- Google AI Pro (about 20 US dollars a month): full access to the Pro model, a large context window, and a multi-terabyte storage allowance (the official page lists 5 TB), sold as part of a Google One plan.
- Google AI Ultra (100 to 200 US dollars a month): the highest limits, the most capable models, top creative tools, and a large storage allowance.
xAI Grok
Built into the X (formerly Twitter) platform and also available at grok.com. Runs the Grok 4.x models and is known for a looser, more conversational style and real-time access to X posts.
- Free: limited access on X and grok.com, basic image generation, and web and X search.
- X Premium (about 8 US dollars a month): bundles basic Grok access inside X.
- SuperGrok (about 30 US dollars a month): full flagship access, larger context, extended thinking, voice, and unlimited image generation.
- SuperGrok Heavy (about 300 US dollars a month): the top tier, with a multi-agent reasoning mode and the largest context window.
Microsoft Copilot
Microsoft's assistant, which runs OpenAI models, making it a Microsoft-branded front end to the same engines behind ChatGPT. Built into Windows and the Office apps.
- Free: basic chat with daily limits and web search.
- Copilot Pro (about 20 US dollars a month): priority access to the latest OpenAI models and AI features inside Word, Excel, PowerPoint, and Outlook (with a Microsoft 365 subscription).
Meta AI
Free and built into WhatsApp, Instagram, Facebook, and Messenger, plus a standalone app. Runs Meta's open-weight Llama 4 models and is funded by Meta's wider business rather than a subscription. A premium tier was in limited testing as of mid-2026 but was not broadly available.
Perplexity
An "answer engine" rather than a single-model chatbot. It searches the web and writes a cited answer, and it can route your question to several different frontier models.
- Free: unlimited basic searches plus a small daily allowance of its stronger "Pro" searches and research queries.
- Pro (about 20 US dollars a month): hundreds of Pro searches a day, deep research, file uploads, and a model picker spanning GPT, Claude, Gemini, and Grok in one place.
- Max (about 200 US dollars a month): the highest limits and earliest features.
DeepSeek and Kimi
Two capable apps from Chinese labs, useful to know but with caveats for some users.
- DeepSeek: the website and apps are free with no paid consumer tier, running the DeepSeek V4 models with reasoning and web search. Paid use is developer-API only.
- Kimi (Moonshot AI): a free web and app tier plus paid plans (roughly 19 US dollars a month and up internationally) that unlock the newest model and agent features. Prices and availability differ by region.
04. Key Terms
| Term | Plain meaning |
|---|---|
| Free tier | The no-cost version of a chat app, with daily message and feature limits. |
| Paid tier / subscription | A flat monthly fee that raises limits and unlocks the best models and features. Often around 20 US dollars a month. |
| Usage cap / limit | A ceiling on how many messages or heavy requests you can send in a period before you must wait or upgrade. |
| Context window | How much text the app can hold in mind at once. Higher tiers usually offer larger windows. See context-window. |
| Deep research | A slower mode that searches many sources and writes a longer, cited answer. |
| Agent mode | A mode where the AI takes multi-step actions on your behalf rather than just replying. See agents-and-agentic-workflows. |
| Answer engine | A product (like Perplexity) that searches the web and returns a cited answer, often using several models. |
05. Examples
- Everyday questions, drafting, and summarising. Any free tier handles this. Start with ChatGPT, Claude, or Gemini and see which interface you prefer.
- You keep hitting the free limit. Pay for one tier around 20 US dollars a month on the app you use most. There is little reason to pay for two at once.
- You want cited, up-to-date answers with sources. Use Perplexity, which is built around web search and citations.
- You already pay Google or Microsoft. Check whether a Gemini or Copilot plan is bundled with a subscription you already have before buying a separate one.
- You want it free and inside apps you already use. Meta AI in WhatsApp or Instagram, or the free tiers of Gemini and Copilot, cost nothing.
06. Common Pitfalls / Misconceptions
"The 200-dollar tier is the one I should buy."
Almost no individual needs it. The power-user tiers (100 to 300 US dollars a month) are for heavy professional use. Most people are well served by a free tier or one 20-dollar plan.
"Paid means private."
Paying does not automatically stop a company from using your conversations to improve its models. That is controlled by a separate setting in each app. Check the privacy or data controls, and turn off training use if you want it off.
"They are all basically the same model."
They run different models with different strengths. Claude is often preferred for writing and careful reasoning, Gemini for tight integration with Google and large documents, Perplexity for sourced web answers, Grok for real-time X content. For most everyday tasks the differences are small.
"The model version names are stable."
They change constantly (GPT-5.x, Gemini 3.x, Grok 4.x, Llama 4, Kimi K2.x, DeepSeek V4). Treat any specific version label as a snapshot. The pricing tiers change more slowly.
"A free chat app and a free local model are the same kind of free."
They are not. A free chat app is a hosted product with limits, run on the company's servers. A free local model runs on your own machine with no company involved.
See running-llms-locally.