Best AI agents: 8 you can use today, from $0 to $25

In brief

Claude Pro at $20 a month is the best single buy for AI agents: it includes Cowork, Claude in Chrome and Claude Code. ChatGPT Plus at $20 adds Work and Codex, and Google AI Pro at $19.99 adds Gemini Tasks. Seven of the eight agents here start at $20 or less, as of November 2026.

What are the best AI agents for personal use? Eight, and what each costs

The best AI agent is the one built into a plan you already pay for, because seven of the eight start at $20 a month or less. We read each maker's pricing and help pages on 15 November 2026. The eight are ChatGPT Work, Claude Cowork with Claude in Chrome, Gemini Tasks, Manus, Genspark, Perplexity Comet, Claude Code and Codex. Each one plans its own steps, uses tools and keeps going without you typing every move.

A chrome cursor arrow, a silver OpenAI knot, a terracotta Claude starburst and a rainbow glass Gemini star, each on its own black plinth, with a row of small glass keys in front of the knot
AgentCheapest plan, November 2026A task it does wellWhere it stops
ChatGPT WorkPlus $20 for the cloud browser; Free gets limited desktop WorkResearches, then fills a form or prepares a booking for your approvalFree and Go get no cloud browser; sites can block it
Claude Cowork and Claude in ChromePro $20Turns messy files and research into spreadsheets and slidesChrome only; Claude Desktop must stay open for local files
Gemini TasksGoogle AI Pro $19.99Watches flight and hotel prices and alerts youNot in the EEA, UK, Switzerland or Nigeria; ages 18 and up
ManusFree, 300 credits a day; Pro from $20Built a 15 minute scoring chart for 200 creditsCan't tell you a task's credit cost beforehand
GensparkFree, 100 credits a day; Plus $24.99Claw acts across email, calendar and messagingLocal Claw only runs while your computer is on
Perplexity CometPro $17 a month billed yearlyResearches the web and organizes your email in a browserFree gets no agent queries; Pro's count isn't published
Claude CodePro $20Writes tests for a module, runs them and fixes failuresNot on the Free plan
CodexFree, limited; CLI on Plus $20Reads a repo, runs commands and reviews changesA Plus window holds about 5 to 45 Astra messages
Seven of the eight AI agents start at $20 a month or less
Perplexity CometPro, billed yearly$17Gemini TasksGoogle AI Pro$19.99ChatGPT WorkPlus$20Claude Cowork and ChromePro$20Claude CodePro$20Codex CLIPlus$20ManusPro, from$20GensparkPlus$24.99each square is one dollar a month on the cheapest paid plan that has theagent, billed monthly where listed (Perplexity lists only yearly), pages read 15 Nov 2026

Claude Pro ($20 a month, or $17 a month billed yearly) is the only plan in the table at that price that carries three named agents, per Anthropic's pricing page and its Cowork and Chrome help pages. ChatGPT Plus has two, Work and Codex. Google lists Tasks, Chrome auto browse and Antigravity on Google AI Pro, which makes it the third bundle worth comparing. Business agent builders such as Zapier Agents, Lindy, Devin and Salesforce Agentforce are outside this list, which covers agents one person can start on a personal plan. Our what is an AI agent guide explains what separates an agent from a chatbot.

Is ChatGPT Work the best AI agent for tasks on the web?

ChatGPT Work is the best pick for tasks that end in a form or a booking, because its cloud browser runs on its own remote computer. OpenAI's help page says it can "read web pages, click buttons, enter information into forms, and carry out steps," and that tasks continue after you close the chat. It starts at Plus, $20 a month, on OpenAI's pricing page.

A task it does well: "Find the cheapest direct flight from Toronto to Lisbon on 12 March and fill in the booking form up to payment. Stop before you submit." ChatGPT asks for confirmation before payments and bookings, and your passwords never go through the chat. If you still need the itinerary itself, our pick of the best AI for travel planning compares the planners. The limits are real. OpenAI's cloud browser page excludes Free and Go, and it warns that "Some websites use security measures that restrict access from automated browser agents." It also doesn't use your own browser's tabs or saved passwords. OpenAI introduced Work on 9 July 2026 as its agent for longer tasks. The older agent mode help page was still live on 15 November 2026, with caps of 40 messages a month on Plus and 400 on Pro. Our agent mode guide has the steps and the 5-hour usage window.

What can Claude Cowork and Claude in Chrome do on the $20 plan?

Claude Cowork turns an outcome into finished files, and Claude in Chrome does the clicking on sites you have open. Both are on Pro, Max, Ultra, Team and Enterprise, per Anthropic's help pages. Cowork runs in the cloud and handles jobs like organizing files, synthesizing research and building spreadsheets with working formulas. Claude in Chrome can read, click, type and fill forms, and it can record a workflow and replay it later.

The limits are specific. Claude in Chrome works only in Chrome, not other browsers and not phones. Anthropic says it is still risky even with its safety classifiers, and it advises against using the built-in browser for financial or medical tasks. Cowork's local files and browser need the Claude Desktop app open. Multi-step tasks use more of your usage than a quick question, so a busy day can hit the cap. Claude Code vs Cowork shows which one fits your job, and is Claude Cowork free covers the plan question.

Is Gemini Tasks worth it, and what happened to Spark?

Gemini Tasks is worth it if you want an agent that watches things while you sleep, and it's what Google renamed Spark after the beta. Google's help page says Gemini can "monitor and alert you when the prices of flights and hotels drop within your budget," and it can also summarize or archive newsletters and package research into PDFs, spreadsheets and presentations. It needs Google AI Pro at $19.99 a month or Ultra, per Google's plans page.

Check the fine print before you pay. You must be 18 or over and signed in with a personal Google account, not a work or school one. Tasks is off in the European Economic Area, Nigeria, Switzerland and the United Kingdom, and it lives in the Gemini app, the Mac app and gemini.google.com. Google calls it experimental. Its warning is blunt: "Remember that Gemini can make mistakes and do unexpected things." A scheduled task can also run while you're offline, when you can't stop it. Our Gemini Spark guide has the rename and the setup.

Is Manus a good AI agent, and is it free?

Manus is a good agent for self-contained builds like charts, web pages and small apps, and it has a free plan. The Free plan gives 300 daily credits, one task at a time and two scheduled automations, per Manus's pricing article. Pro starts at $20 a month for 4,000 credits, and Team starts at $20 a seat.

Manus's credits page lists three real tasks. A 15 minute NBA scoring chart cost 200 credits, a 25 minute wedding webpage cost 360, and an 80 minute sky-events web app cost 900. Those are the only prices you get. Manus's help center says it can't estimate a task's credits before it starts, and it suggests splitting big jobs into smaller ones. That makes a bill harder to plan than a flat plan. Is Manus free works through what the free credits buy.

Where does Genspark beat the big three agents?

Genspark beats them at acting across your messaging apps. Its help center describes Claw as a personal assistant that can "take action for you across email, calendar, messaging, and more," with 30 or more built-in skills including phone calls, and you can reach it from WhatsApp, Slack or Telegram. Super Agent is the other half: a session-based assistant on genspark.ai that starts fresh each time.

Plans come from Genspark's membership page, updated 3 November 2026. Free gives 100 credits a day up to a lifetime limit, Plus is $24.99 a month ($19.99 a month billed yearly) for 10,000 credits, and Pro is $249.99 for 125,000. The catches: every message, tool call and scheduled run spends credits, local Claw only runs while your computer is on with the app open, and the always-on cloud version needs a separate Cloud Computer subscription. Genspark also warns about prompt injection from untrusted sites and emails. Genspark pricing has the credit math.

Is Perplexity Comet the best AI agent for research?

Perplexity Comet is the best fit if you want research and the agent in one browser, but you'll pay to use the agent. Perplexity's Comet page says it can automate tasks, research the web and organize your email, on Mac, Windows, iOS and Android. Its plan comparison lists browser agent queries as not included on Free.

Pro is $17 a month billed yearly on Perplexity's pricing hub, and it gets monthly limits for average use with no number published. Max is $200 a month, and Perplexity says it has the highest weekly limit on browser agent queries. Only the Enterprise plans publish a count: 80 a month on Enterprise Pro and 800 on Enterprise Max. So a heavy user can't tell on Pro how many tasks the plan holds.

What is the best AI agent for coding, Claude Code or Codex?

Claude Code and Codex are the two best coding agents you can start on a $20 plan, and they differ most on price at the bottom. Claude Code needs a paid Claude plan or an Anthropic Console account, so Free is out. Codex is included in ChatGPT Free, Go, Plus, Pro, Business, Edu and Enterprise, per OpenAI's pricing page, with Free limited to quick tasks. Codex CLI access starts at Plus. Claude Code vs Codex compares them task by task.

Each one installs with one command from its maker's docs. For Claude Code on macOS, Linux or WSL, Anthropic's overview gives curl -fsSL https://claude.ai/install.sh | bash, and irm https://claude.ai/install.ps1 | iex for Windows PowerShell. Then run claude in a project folder. For Codex, OpenAI's CLI page gives curl -fsSL https://chatgpt.com/codex/install.sh | sh, and the first run asks you to sign in with ChatGPT. Anthropic's own example prompt is claude "write tests for the auth module, run them, and fix any failures".

The limits matter more than the install. OpenAI's page estimates a Plus window at 5 to 45 messages on GPT-6 Astra and 350 to 3,000 on GPT-6 Luna, and it says those aren't fixed limits. Pro plans have no five-hour limit at the moment. Neither agent should run on code you can't review. Claude Code pricing and Codex pricing cover what the bigger plans add.

How do you pick the right AI agent for your job?

Pick by the job, then by the plan you already hold. Four rules cover most readers, and they take a minute to apply:

  1. Your work ends in a web form or a booking: use ChatGPT Work, and keep its confirmation prompts on.
  2. Your work ends in a spreadsheet, a deck or a folder of tidy files: use Claude Cowork.
  3. You want something to watch prices or inboxes in the background: use Gemini Tasks if your country is on its list, and Genspark Claw if you live in messaging apps.
  4. You write code: start with Claude Code or Codex, whichever maker you already pay.

Three prompts work on nearly any of them. Each names the goal and the stop line, which is the part people forget:

  1. "Research the five best-reviewed standing desks under $500, put them in a table with price, warranty and return window, and cite each page. Do not buy anything."
  2. "Open my downloads folder, group the files by project, and rename them with the date first. Show me the plan before you move anything."
  3. "Every morning, check the price of flight ABC123 and tell me if it drops below my budget. Never book."

Every agent in this guide is paid or credit-limited past a small free allowance, so test one before you commit to a year. Our is Manus free and Genspark pricing pages show what a free allowance buys.

What if you want answers from Claude, GPT and Gemini, not an agent?

If you only need answers, one app that holds all three assistants costs less than the agent plans above. Whizi doesn't run any of the eight agents. It has no Work, no Cowork, no Tasks, no Claude Code and no Codex, and no browser that clicks for you. What it does carry is Claude, GPT and Gemini in one app on Pro, with the first week at $1.99, and web search on by default for each new chat.

The free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5, with no card. That fits the reading half of agent work: comparing options, drafting the email, checking one model's answer against another. Use an agent for the clicking half. A Project can hold your instructions and files for every chat in it.

Is it safe to let an AI agent act for you?

It's safe for low-stakes, reversible tasks and risky for money, health and anything you can't undo. Every maker above asks for confirmation before payments and bookings, and every one says it can be wrong. The main attack is prompt injection, where text on a web page or in an email tells the agent to do something you didn't ask. OpenAI, Anthropic, Google and Genspark each describe it in their own docs.

Four habits cut the risk. Name the stop line in your first message. Keep the confirmation prompt on. Type passwords into the secure sign-in form, never into the chat. Don't schedule a task you wouldn't want running while you're away, since Google notes you may be unable to stop a scheduled task that runs offline.

Workflow checklist
  • Start with the agent in the plan you already pay for: Claude Pro or ChatGPT Plus at $20, or Google AI Pro at $19.99
  • Say the stop line in the first message, such as "prepare the booking, do not submit"
  • Keep confirmation prompts on until a task has run cleanly three times
  • Type passwords and card numbers into the secure form or your own browser, never into the chat
  • Try Manus or Genspark on their free daily credits before paying for either
  • Check Gemini Tasks availability for your country before you buy Google AI Pro for it
Common questions

Frequently asked questions

What is the best free AI agent?

Manus has the most generous free agent: 300 credits a day, one task at a time and two scheduled automations, as of November 2026. Genspark gives 100 credits a day up to a lifetime limit, and Codex has a limited free tier for quick coding tasks. Is Manus free shows what the credits buy.

What is the best AI agent for coding?

Claude Code and Codex are the two to start with, and both begin at $20 a month on Claude Pro or ChatGPT Plus. Claude Code has no Free plan, while Codex has a limited free tier. Pick the one whose maker you already pay. Claude Code vs Codex compares them.

Is ChatGPT agent still available?

Yes, in two forms, as of 15 November 2026. OpenAI's agent mode help page is still live with monthly caps of 40 on Plus and 400 on Pro, and its release notes show no retirement. Work, introduced on 9 July 2026, is the newer agent, and its cloud browser needs a paid plan, not Free or Go. How to use agent mode in ChatGPT has the details and the usage limits.

Can an AI agent book flights and make purchases for me?

It can prepare them, and it pauses for your approval before paying. ChatGPT Work and Gemini Tasks both ask for confirmation before payments or bookings, and neither wants card numbers typed into the chat. Give it a stop line like "do not submit" and finish the payment yourself.