Which AI does what, exactly?
Everyone mentions ChatGPT, Claude, Gemini, open source, agents… and almost nobody explains what any of it means. This is your plain-English guide: no benchmarks, no jargon, no assumption that you’ve been following the AI industry for years.
“I spent a solid three years confused about which AI I was supposed to be using. Turns out the answer is: it depends. But now at least I can tell you what it depends on.”
The Big Names (Frontier Models)
The most powerful and widely-used AI tools: the ones your friends, family, and the news are almost certainly talking about. Made by large companies, mostly with free tiers. If you're just starting out, this is where most people begin.
Not sure what ‘frontier model’ or ‘AI agent’ means? Kevin has a plain-English glossary. →
The tool that started the conversation, and still one of the best for almost anything you ask it to do.
Free tier available · Paid plans for more usage.
No — proprietary (OpenAI keeps this to themselves)
Particularly strong at reading long documents, explaining things clearly, and tasks that need careful reasoning.
Free tier available · Paid plan for more usage.
No — proprietary
Google's AI. Works across Search, Docs, and Gmail, and genuinely useful if you're already in the Google ecosystem.
Free tier available · Advanced plan via Google One subscription.
No (Gemma — Google's open version — is though)
Built into Windows and Microsoft 365. If you live in Word and Excel, Copilot is already waiting.
Free version built into Windows & Edge · Paid plan for Microsoft 365 integration.
No — proprietary
Real-time access to X. Useful when you need an AI that's actually caught up with today's news.
Free with an X (Twitter) account · Paid plan for more usage.
No — proprietary (built by xAI)
One app, many chatbots. A simple way to try ChatGPT, Claude, and Gemini side by side without juggling separate logins.
Gives you access to many different chatbots — ChatGPT, Claude, Gemini, Grok and more — through one app and one subscription.
Free tier available · Paid plan for heavier use and premium bots.
No — proprietary (made by Quora)
Open Source & Local Models
These models have their code — and often the AI itself — freely available to download. Some can run entirely on your own computer so nothing leaves your machine. They take a bit more effort to set up, but they're free and extremely private.
“The code is public. The AI is often free. For some people, that is the whole point.”
The engine behind many AI apps you've heard of: powerful, open, and free to use without a subscription.
Free — download directly or use via Meta AI.
Yes — fully downloadable and free to use
A lightweight European model that handles writing and summarisation well, without the data concerns of US or Chinese alternatives.
Free to download. Also available as a hosted chatbot.
Yes — and it's a European company, so subject to EU data laws
Surprised the entire AI industry with its reasoning ability, though it routes through Chinese infrastructure, which is worth weighing.
Free.
Yes — ⚠️ built and operated in China. Use caution with sensitive personal information.
Microsoft's compact model, capable enough for most tasks and designed to run on more modest hardware.
Free — specifically designed to run well on less powerful computers.
Yes
Google's free, open model family. The small, downloadable cousin of Gemini, made to run on your own hardware.
Free — open and downloadable.
Yes — open weights, free to download and run
AI Agents & Specialist Tools
These go beyond 'ask a question, get an answer.' AI agents carry out multi-step tasks on your behalf — searching, reading, comparing, and presenting results. You're hearing about these constantly right now because they represent a genuinely new way of working.
“This is the part that surprises people. You stop asking. You start delegating.”
Every answer comes with sources you can click and check — the AI tool that shows its working.
Free tier available · Paid plan for more daily searches.
No — proprietary
Upload your own documents and ask questions about them, like having a researcher read everything for you.
Free.
No
Connects AI to your browser so it can read pages, fill in forms, and complete web tasks on your behalf.
A personal AI agent that lives on your computer and you talk to through apps you already use — WhatsApp, Telegram, Discord. It manages email, calendar, reminders, and background tasks on your behalf, 24/7.
Free — open source. You do need a subscription to an AI model (like Claude or ChatGPT) for it to actually run.
Yes — your data and skills stay on your own computer, not a company's server
⚠️ Requires setup — this is not a click-and-go tool. Very new (launched early 2025). Very popular, very fast-moving.
Give it a complex question, wait a few minutes, and get back a properly sourced report instead of a quick answer.
A research agent that searches dozens of live sources and returns a detailed written report — with every claim clearly sourced. Perplexity's strength has always been showing exactly where information comes from, and that carries through here.
Paid — part of Perplexity Pro subscription.
An open-source deep research agent you can run yourself, with full control over what it reads and how it works.
An open source research agent — similar to Deep Research but free to download and run yourself, with full control over how it works.
Free — open source framework. Technical setup required.
⚠️ Requires technical setup — not a click-and-go tool
Google's research agent: searches the full web, reads multiple sources, and writes you a structured summary.
Google's answer to Deep Research — searches across the web and your connected Google documents, then produces a structured report.
Free tier available · Full access with Gemini Advanced subscription.
Can search your Google Drive and Docs as part of its research — unique to Google's ecosystem
An open-source personal agent that lives on your computer, remembers your work, and reaches you through Telegram, Discord, or Slack.
A personal AI agent that runs on your own computer, keeps a long-term memory of your projects, and reaches you through messaging apps. It can run scheduled tasks, research, and routine jobs on your behalf.
Free — open source. You'll need a subscription to an AI model (like Claude or ChatGPT) for it to run.
Yes — your data and skills stay on your own machine
⚠️ Requires setup — built for confident, technical users. Very new and fast-moving.
Run AI on Your Own Computer
These aren't AI models themselves — they're the apps that let you run open source models locally, entirely on your own machine. Nothing leaves your computer. No subscription, no internet required once set up. The difference between the two is mainly about how comfortable you are with clicking buttons versus typing commands.
“No internet. No company logging your questions. Just you and a very fast computer.”
The friendliest way to run AI on your own computer: a proper app with a simple interface, no commands needed.
A proper desktop app — looks and feels like real software — that lets you browse, download, and chat with local AI models. The most beginner-friendly way to run AI locally. No command line needed.
Open it like any other app, browse a catalogue of models, click download, and start chatting. Has a built-in chat interface that looks a lot like ChatGPT — but everything runs on your machine.
Free for personal use. The models it runs are also free.
Mac, Windows, or Linux. At least 8GB RAM — 16GB recommended. No technical skills required.
For those comfortable with a terminal: run powerful open models locally with complete privacy and no subscription.
A lightweight tool that runs in the background and lets you pull and run AI models with simple typed commands. More technical than LM Studio — but extremely popular, and the engine that OpenClaw runs on top of.
You type a short command and the model downloads and runs. Supports 100+ models including Llama, Mistral, Qwen, Phi-4, and DeepSeek. Pair it with a visual front-end (like LM Studio) if you prefer buttons over commands.
Free — and so are most of the models it runs.
Mac, Windows, or Linux. 8GB RAM minimum. Some comfort with a terminal or command prompt helps — though the basics are very simple.
The lean open-source engine that quietly runs AI models on ordinary computers. Powerful, free, and not for the faint-hearted.
A lightweight, open-source engine for running AI models on everyday computers — even without a high-end graphics card. It's the foundation many friendlier local-AI apps are built on.
Free — open source.
Yes — fully open
Mac, Windows, or Linux. Comfort with a command line helps — most people use it through a friendlier app instead.
A newer way to run AI models fast on Apple Silicon Macs, built to make the most of Mac hardware. Mac-only, and still early.
A local AI runner built specifically for Apple Silicon Macs, using Apple's MLX framework to run models faster and manage memory more efficiently than general-purpose tools.
Free — open source.
Yes
An Apple Silicon Mac (M1 or newer). Mac-only.
Apple's own open-source framework for running AI on Mac hardware. The foundation, more for builders than for browsers.
An open-source framework from Apple for running AI efficiently on Apple Silicon. It's a foundation that other Mac AI tools are built on, aimed at developers more than everyday users.
Free — open source.
Yes — made by Apple
An Apple Silicon Mac. Built for developers.
The internet's library of open AI. Where the world's free models and datasets live, and where much of open AI is shared.
A huge public platform hosting hundreds of thousands of free, open AI models and datasets — the central hub of the open AI world.
Free to browse and download · Paid plans for hosting and heavier compute.
The platform hosts open models; many are fully open
The Global Field
AI is no longer just a Silicon Valley story. China in particular has produced several world-class models that rival — and in some cases lead — the Western field. These come up constantly in tech conversations and news, so it's worth knowing they exist. A note worth making: all three listed here are from Chinese companies. That's worth being aware of when it comes to what data you share with them.
“Worth knowing they exist. Worth knowing where they're from. Worth making that call yourself.”
Alibaba's model, strong in both Chinese and English and one of the most capable widely-used Chinese-origin AI tools.
Many versions are free and open source. Hosted version available via Alibaba Cloud.
Many versions yes — downloadable and free to run. Made by Alibaba, the Chinese e-commerce giant. ⚠️ Use caution with sensitive personal data.
A free-to-try chatbot from China’s Moonshot AI. The K3 model handles enormous context, reads images, and is useful when smaller tools run out of room.
Kimi K3, released July 2026: a mixture-of-experts model with about 2.8 trillion total parameters and roughly 104 billion active per token.
Up to 1 million tokens. That is the reason to try it when you have long PDFs, large notes, or a codebase that makes smaller chatbots start politely losing the plot.
Long-document research, coding help, agent-style browsing and automation tasks, and working with text plus images in the same conversation.
Start at kimi.com for the consumer chatbot, or use platform.kimi.ai for the API. Developers can call the Kimi K3 model through an OpenAI-compatible SDK path.
Launch pricing was reported at about $0.30 per million cached input tokens, $3 per million fresh input tokens, and $15 per million output tokens. Check Moonshot's live pricing before budgeting anything serious.
Moonshot has published full Kimi K3 weights on Hugging Face under the Kimi K3 License. Open weights do not mean laptop-friendly: this is a very large model, not a weekend download for a normal home setup.
Moonshot AI is based in China. If privacy, data residency, or training-on-your-chats matters for what you are uploading, treat that as a separate privacy check instead of assuming U.S. defaults.
A capable Chinese model for both conversation and longer writing tasks, with a growing international user base.
Free to use on their apps · API access available.
M2 series yes — downloadable. ⚠️ Chinese company — worth noting for privacy-conscious users.
Making Images & Video
These tools turn a plain description into a picture or a video clip. Type what you want to see and they produce it: illustrations, photos of things that never happened, even short cinematic scenes. Each one has its own strengths and its own rules about how you can use what it makes, so check the profiles before publishing anything.
“You type what you want to see. It makes it. I still find that slightly unreasonable.”
The image generator known for sheer beauty. If you want pictures that look like art, this is the one people reach for.
Creates striking, often beautiful images from text descriptions. Known for its artistic quality.
Paid — subscription required, no free tier.
No — proprietary
OpenAI's image generator, built right into ChatGPT. Describe a picture in plain words and it appears, no separate tool needed.
Generates images from text descriptions, built directly into ChatGPT. Just ask, and it draws.
Free tier available · More on paid ChatGPT plans.
No — proprietary (OpenAI)
The open image generator that runs on your own machine. Free, private, and endlessly customisable, if you're willing to tinker.
Generates images from text, and can run entirely on your own computer. Free and open, with a large community of custom add-ons.
Free — open and downloadable.
Yes — free to download and run
Adobe's image and video generator, built to be commercially safe. Made for people who need to use the results in real work.
Generates images and video from text, designed so the results are safe to use commercially. Built into Adobe's apps.
Free tier available · Included in Adobe subscriptions.
No — proprietary (Adobe)
A pioneer of AI video. Turns text and images into short film clips, and helped start the whole AI-video wave.
Creates short video clips from text prompts or still images. One of the longest-running AI video tools.
Free tier available · Paid plans for more and longer videos.
No — proprietary
A Chinese video generator known for smooth, realistic motion. Capable and popular, with the usual data caveats.
Generates realistic short video clips from text or images, known for smooth, natural motion.
Free tier available · Paid plans for more usage.
No — proprietary
xAI's image and video generator, built into Grok. Fast, loosely filtered, and aimed at quick creative experiments.
Creates images and short videos with sound from text or photos, built into Grok. Built for speed and quick iteration.
Free tier available · More via X / Grok subscription.
No — proprietary (xAI)
ByteDance's video generator, from the company behind TikTok and CapCut. Strong, cinematic clips, with the usual data caveats.
Generates cinematic short videos with synchronised audio from text, images, or reference clips. Made by ByteDance, accessed through Dreamina and CapCut.
Free daily credits · Paid for more.
No — proprietary
Google's flagship video generator. Among the most realistic AI video available, and now free to try through Google.
Generates highly realistic short videos with sound from text or images. Google's flagship video model, accessed through Gemini and Google Flow.
Free tier for Google account holders · More on paid plans.
No — proprietary (Google)
The design app most people already know, now with AI woven throughout. Familiar, friendly, and quietly very capable.
A friendly design app for posters, social posts, presentations and more, now with AI built in for generating images, writing text, and editing.
Free tier available · Paid plan for premium features.
No — proprietary
Music, Voice & Sound
AI for the things you hear. Some of these write and perform complete songs from a text description; others turn writing into natural-sounding speech, in almost any voice and language. A small section for now, but it covers some of the most surprising things AI can currently do.
“I made a song about my printer. It had a chorus. We live in remarkable times.”
The leading AI voice tool. Turns text into remarkably natural speech, and can clone a voice from a short sample.
Turns written text into natural-sounding speech, and can recreate a specific voice from a short sample. Used for narration, audiobooks, and dubbing.
Free tier available · Paid plans for more audio and features.
No — proprietary
Write a song by describing it. Suno turns a few words into a complete track, vocals and all, in under a minute.
Creates complete songs — music and vocals — from a short text description. Pick a style, add lyrics if you like, and it writes the rest.
Free tier available · Paid plans for more songs and commercial use.
No — proprietary
Google’s browser music studio: type what you want, get a full track with vocals, and steer it like you’re talking to a patient producer.
A free-to-start web app from Google that turns text, image, and audio prompts into songs using DeepMind’s Lyria 3.5 music model. Google launched Lyria 3.5 in Flow Music on 29 July 2026, with claims of better musicality, lyrics, vocals, and control.
Beginner-friendly if you can use a browser and a Google account. No DAW or music theory required. You prompt a Producer-style chat, then edit, remix, share, or download the song.
Free plan: $0, daily credit top-ups, 2 concurrent generations, and core features. Pricing page checked 2026-08-10 showed annual-billing rates of Starter $6/month, Plus $18/month, and Member $48/month, with about 600, 2,000, and 6,000 songs respectively. Monthly billing may show higher numbers, so check the live toggle before subscribing.
Downloadable songs; the pricing comparison lists mp3, wav, and m4a downloads, plus stem downloads. The product surface also advertises projects, sharing, and AI music videos using Veo. Outputs carry Google’s SynthID watermark, which is provenance labeling, not a separate ban on use.
Google says it does not claim ownership of original content generated with Flow Music, and Google’s general terms say original generated content remains yours. That does not erase rights questions around uploads, artist impersonation, commercial use, or music-platform rules. Treat commercial use as plan-and-terms dependent, and read the current terms before using a track in paid work.
Flow Music’s privacy notice says Google collects uploads, generated outputs, usage information, and feedback, and may use them to improve Google products and machine-learning technologies. Human reviewers may process a sample, with account disconnection steps. Do not upload confidential audio, private family material, or anything you would not want reviewed.
Quality varies by prompt; expect some generic AI-band moments. This is not a replacement for a human collaborator or a traditional DAW workflow. Rights, training data, and platform rules in AI music remain contested industry-wide. The card is a product map, not legal advice.
Same basic job: make a song from a sentence. Flow Music is Google’s version, with a Lyria engine, a Google account, and optional video tools. Suno is the dedicated music-product competitor already in the Lineup. Pick based on which one sounds better to your ears, not brand loyalty. Old-fashioned, I know.
Building Apps & Websites
These use AI to build software: websites, apps, and small personal tools. Some are editors made for professional programmers, but the newest ones let you describe what you want in plain English and watch it take shape. You don't need to read a line of code to end up with something that works.
“Twenty-two years in IT, and now the computer writes the code. I've decided to find that relaxing.”
An AI-first code editor that's become the developer favourite. For people who write code and want AI deeply built in.
A code editor with AI built into its core — it can write, explain, and fix code as you work. Aimed at people who write software.
Free tier available · Pro plan for heavier use.
No — proprietary (built on the open VS Code)
Some coding background. This is a developer tool.
An AI code editor now owned by the team behind Devin. Powerful, fast-moving, and recently rebranded, so the name may shift.
An AI-powered code editor, now owned by Cognition (makers of the Devin coding agent) and being rebranded toward 'Devin Desktop.' For people who write software.
Free tier available · Pro plan for more.
No — proprietary (built on the open VS Code)
A coding background. This is a developer tool, and its branding is currently changing.
Describe an app in plain English and watch it get built. The poster child for 'vibe coding,' aimed at non-coders with an idea.
Builds working web apps from a plain-language description — handling the design, the logic, and the database. No coding required.
Free tier available · Paid plans for serious building.
No — proprietary
Just an idea and plain words. The free tier is enough to test one.
Hand Off the Busywork (Automation)
These connect the apps you already use, like email, spreadsheets, and calendars, and quietly handle the repetitive steps between them. "When an invoice arrives, save the attachment and add a row to my spreadsheet." You set the rule up once, and it keeps happening without you.
“It does the boring parts while you do something better. The closest thing yet to a second Kevin.”
A powerful automation tool you can run yourself. Connects your apps and lets them work together, with full control over your data.
Connects your apps and automates tasks between them — 'when an email arrives, save the attachment and log it.' Can run on your own server for full data control.
Free to self-host · Paid cloud version for convenience.
Source-available (fair-code) — free to self-host; not strictly open source
⚠️ More setup than Zapier or Make, especially self-hosted
The original app-connector. Links thousands of services together so routine tasks happen on their own, no code needed.
Connects thousands of apps so routine tasks happen automatically — no code, just pick a trigger and an action.
Free tier available · Paid plans for more automations.
No — proprietary
Beginner-friendly — built for non-technical users
A visual automation tool where you build workflows by connecting dots on a canvas. More flexible than Zapier, a little more involved.
Connects apps and automates tasks, built around a visual canvas where you join steps together. More flexible than the simplest tools.
Free tier available · Paid plans for more operations.
No — proprietary
A little more involved than Zapier, but very capable
This guide is updated quarterly. Tools change fast — if you spot something out of date, let us know.