本栏目尚未翻译。以下内容按原文以英文显示。 阅读英文版本.
Last reviewed · How we check sources
AI agents · updated 1 Nov 2026
A model answers. An agent acts: it plans, uses tools, checks the result and keeps going. Use the finder to shortlist 42 agents in seven categories, from coding agents and browser agents to enterprise platforms, open-source frameworks, voice and research, by your task, budget, skill, deployment and privacy needs. Every value links to the maker's own page. For the models underneath, see AI models compared.
AnswersHelps a personReplies automaticallyPlans and acts with tools
Plain-language definitions by CoolPower Index. The line between them is blurring: many products now combine an assistant with agent features.
42
agents and frameworks
26
makers
24
support MCP
29
have a free plan
15 of 42 agents match. Filtered out: 27 built for other jobs.
Google · Coding · Developer
Why: Matches your task. A free agent-first editor.
Free · free tier antigravity.google (opens in a new tab)
Anysphere · Coding · Developer
Why: Matches your task. An AI-first editor with agent and tab completion in one.
From US$8 a month · free tier · MCP cursor.com (opens in a new tab)
xAI · Coding · Developer
Why: Matches your task. People who already pay for an X subscription.
From US$8 a month · MCP x.ai (opens in a new tab)
OpenAI · Coding · Developer
Why: Matches your task. Delegating many parallel coding tasks from ChatGPT.
From US$8 a month · free tier · MCP · open source developers.openai.com (opens in a new tab)
GitHub (Microsoft) · Coding · Developer
Why: Matches your task. Teams already on GitHub.
From US$10 a month · free tier · MCP github.com (opens in a new tab)
Your task, must-haves, budget, skill and deployment remove agents that do not fit. The rest get points: 3 when the agent is built for your task, 2 when it suits your skill level, 1 when it offers the autonomy you asked for and 1 for a free tier when you choose Free only. Ties are ordered by lowest published starting price, then name. Every attribute comes from the maker's own pages, linked in Details. We do not rate quality.
Pick two to four agents, here or with the Compare box on a result card or table row.
| Agent | Maker | Category | From | Free tier | Runs on | MCP | Open source | Compare |
|---|---|---|---|---|---|---|---|---|
| Anthropic | Coding | From US$20 a month claude.com (opens in a new tab) | No | Your computer, Cloud | Yes | No | ||
| OpenAI | Coding | From US$8 a month developers.openai.com (opens in a new tab) | Yes | Your computer, Cloud | Yes | Yes | ||
| GitHub (Microsoft) | Coding | From US$10 a month github.com (opens in a new tab) | Yes | Your computer, Cloud | Yes | No | ||
| Anysphere | Coding | From US$8 a month cursor.com (opens in a new tab) | Yes | Your computer, Cloud | Yes | No | ||
| Cognition | Coding | From US$20 a month devin.ai (opens in a new tab) | Yes | Your computer, Cloud | – | No | ||
| Replit | Coding | From US$20 a month replit.com (opens in a new tab) | Yes | Cloud | – | No | ||
| Lovable | Coding | See note lovable.dev (opens in a new tab) | Yes | Cloud | Yes | No | ||
| Amazon (AWS) | Coding | From US$20 a month kiro.dev (opens in a new tab) | Yes | Your computer | – | No | ||
| Coding | Free antigravity.google (opens in a new tab) | Yes | Your computer | – | No | |||
| xAI | Coding | From US$8 a month x.ai (opens in a new tab) | No | Your computer | Yes | No | ||
| Meta | Coding | See note dev.meta.ai (opens in a new tab) | No | Your computer | – | No | ||
| All Hands AI | Coding | From US$500 a month all-hands.dev (opens in a new tab) | Yes | Your computer, Self-host, Cloud | Yes | Yes | ||
| Cline | Coding | See note cline.bot (opens in a new tab) | Yes | Your computer | Yes | Yes | ||
| Anthropic | Browser and computer use | From US$20 a month claude.com (opens in a new tab) | No | Your computer | – | No | ||
| Anthropic | Browser and computer use | See note docs.anthropic.com (opens in a new tab) | No | Self-host | – | No | ||
| Browser Use | Browser and computer use | See note browser-use.com (opens in a new tab) | Yes | Cloud, Self-host | Yes | Yes | ||
| Manus AI | Browser and computer use | From US$20 a month help.manus.im (opens in a new tab) | Yes | Cloud | – | No | ||
| Perplexity | Browser and computer use | Free perplexity.ai (opens in a new tab) | Yes | Your computer | – | No | ||
| xAI | Assistant with agent mode | See note x.ai (opens in a new tab) | No | Cloud | Yes | No | ||
| OpenAI | Assistant with agent mode | From US$8 a month developers.openai.com (opens in a new tab) | Yes | Cloud | Yes | No | ||
| Anthropic | Assistant with agent mode | From US$20 a month claude.com (opens in a new tab) | Yes | Cloud, Your computer | Yes | No | ||
| Assistant with agent mode | From £4.49 a month gemini.google (opens in a new tab) | Yes | Cloud | – | No | |||
| Microsoft | Assistant with agent mode | From US$30 a month microsoft.com (opens in a new tab) | No | Cloud | – | No | ||
| Microsoft | Enterprise platform | From US$200 a month microsoft.com (opens in a new tab) | No | Cloud | Yes | No | ||
| Salesforce | Enterprise platform | From US$550 a month salesforce.com (opens in a new tab) | No | Cloud | – | No | ||
| Amazon (AWS) | Enterprise platform | See note aws.amazon.com (opens in a new tab) | No | Cloud | Yes | No | ||
| Enterprise platform | See note ai.google.dev (opens in a new tab) | No | Cloud | Yes | No | |||
| Anthropic | Enterprise platform | See note docs.anthropic.com (opens in a new tab) | No | Cloud | – | No | ||
| Zapier | Enterprise platform | From US$19.99 a month zapier.com (opens in a new tab) | Yes | Cloud | Yes | No | ||
| n8n | Enterprise platform | From €20 a month n8n.io (opens in a new tab) | Yes | Cloud, Self-host | Yes | Yes | ||
| OpenAI | Framework or SDK | Free openai.com (opens in a new tab) | Yes | Self-host | Yes | Yes | ||
| Anthropic | Framework or SDK | Free github.com (opens in a new tab) | Yes | Self-host | Yes | Yes | ||
| Framework or SDK | Free github.com (opens in a new tab) | Yes | Self-host, Cloud | Yes | Yes | |||
| LangChain | Framework or SDK | From US$39 a month langchain.com (opens in a new tab) | Yes | Self-host, Cloud | Yes | Yes | ||
| Microsoft | Framework or SDK | Free github.com (opens in a new tab) | Yes | Self-host | Yes | Yes | ||
| CrewAI | Framework or SDK | See note crewai.com (opens in a new tab) | Yes | Self-host, Cloud | Yes | Yes | ||
| Mastra | Framework or SDK | Free github.com (opens in a new tab) | Yes | Self-host, Cloud | Yes | Yes | ||
| ElevenLabs | Voice | From US$6 a month elevenlabs.io (opens in a new tab) | Yes | Cloud | – | No | ||
| Vapi | Voice | From US$29 a month vapi.ai (opens in a new tab) | Yes | Cloud | – | No | ||
| Retell AI | Voice | See note retellai.com (opens in a new tab) | Yes | Cloud | – | No | ||
| Research | See note ai.google.dev (opens in a new tab) | No | Cloud | – | No | |||
| Perplexity | Research | From US$200 a month perplexity.ai (opens in a new tab) | Yes | Cloud | – | No |
Prices are the lowest paid plan per person per month on the maker's page, checked 1 November 2026. A dash means not stated on the pages we checked.
Counts from the table, not a ranking.
Only published prices: model tokens from the AI models table, and Amazon Bedrock AgentCore Runtime compute. Subscription plans are in the finder and table above.
For scale, Google says one managed agent interaction typically uses 100k to 3M tokens. ai.google.dev (opens in a new tab)
Presets from Google's published range, counted as input tokens:
Model tokensUS$1,500.00 developers.openai.com (opens in a new tab)
Model tokens AgentCore compute
Estimated monthly totalUS$1,500.00
Your own inputs times published list prices, before caching, batch or volume discounts. Tools, storage, gateways and memory are billed separately where they apply. Products without a published price are left out.
Pattern described in the A2A documentation: MCP standardises how an agent reaches its tools, A2A how independent agents discover each other and delegate. Diagram by CoolPower Index.
Pick one task with a clear finish line, such as reviewing a pull request or researching a topic. The finder above shortlists agents by job.
A product such as Claude Code works out of the box. A framework such as the OpenAI Agents SDK or Google ADK lets developers build their own. openai.github.io (opens in a new tab)
MCP is an open standard for connecting AI applications to external systems such as files, databases and tools. modelcontextprotocol.io (opens in a new tab)
Hosted agents run in sandboxes. Google provisions a Linux sandbox per call, and AgentCore runs each session in its own microVM. docs.aws.amazon.com (opens in a new tab)
Anthropic warns that prompt injection can reach agents through web pages or images, so check consequential steps yourself. docs.anthropic.com (opens in a new tab)
Agents call the model many times per task. Use the calculator above with your own volumes. ai.google.dev (opens in a new tab)
415 TWh
Data centre electricity use in 2024, about 1.5% of world consumption.
~945 TWh
IEA projection for data centre electricity use by 2030, more than double 2024. AI is the most important driver.
1,000 to 1,800
Input tokens per screenshot in a computer use agent loop, so long agent runs build up tokens quickly.
~4,500
Extra input tokens added to each request just by declaring the computer use toolset.
2030 and 2035 are IEA projections (dashed). Across the IEA's cases 2035 ranges from 700 to 1700 TWh.
Source: IEA, Energy and AI (opens in a new tab)
| Year | TWh | Status |
|---|---|---|
| 2024 | 415 | Estimate |
| 2030 | 945 | Projected |
| 2035 | 1200 | Projected |
100,000 to 3,000,000
One Gemini managed agent interaction, typical
Source: Google docs (opens in a new tab)
Different measures from different makers, shown for scale only. Each screenshot and the tool definitions are sent to the model as input tokens.
An agent makes many model calls per task instead of one, and each call can carry screenshots, tool definitions and earlier results. That multiplies the tokens, and so the accelerator time, behind a single request. The IEA figures cover all data centres, of which AI is the main growth driver. No maker publishes energy per agent task, so we show none. For capacity and grid data see chip power, energy and market intelligence.
An open-source standard for connecting AI applications to external systems: data sources, tools and workflows. Its makers compare it to a USB-C port for AI applications. Supported by Claude, ChatGPT, Visual Studio Code, Cursor and others.
An open standard for agents to talk to other agents: discover each other, delegate tasks and share results. Originally developed by Google and donated to the Linux Foundation. MCP is agent to tool, A2A is agent to agent, and the two are designed to work together.
Quoted from each maker's documentation. We did not run these and show no outputs of our own.
from agents import Agent, Runner
agent = Agent(name="Assistant", instructions="You are a helpful assistant")
result = Runner.run_sync(agent, "Write a haiku about recursion in programming.")
print(result.final_output)interaction = client.interactions.create(
agent="antigravity-preview-09-2026",
input="Audit https://web.dev for performance, Core Web Vitals, and SEO. ...",
environment="remote",
)# check the install
claude --version
# pull a web or mobile session into the terminal
claude --teleport
# start a task in the cloud
claude --cloudimport asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
async def main() -> None:
agent = AssistantAgent("assistant", OpenAIChatCompletionClient(model="gpt-4o"))
print(await agent.run(task="Say 'Hello World!'"))
asyncio.run(main())from langgraph.graph import StateGraph, MessagesState, START, END
def mock_llm(state: MessagesState):
return {"messages": [{"role": "ai", "content": "hello world"}]}
graph = StateGraph(MessagesState)
graph.add_node(mock_llm)
graph.add_edge(START, "mock_llm")
graph.add_edge("mock_llm", END)
graph = graph.compile()
graph.invoke({"messages": [{"role": "user", "content": "hi!"}]})Unconfirmed. Reported by named outlets, not confirmed by the companies. Each links to the report with its date.
Reported to be in internal testing since late June. Cursor has not announced it or confirmed a release. Originally reported by The Information. The Opinion: Cursor preparing to rival Claude's Cowork (opens in a new tab)
Based on people familiar with the work. Anthropic has not announced it. Originally reported by The Information. Investing.com: Anthropic, Google and OpenAI prepare new AI models, The Information reports (opens in a new tab)
Confirmed since: 8 Oct 2026: Google Cloud introduced the Gemini agent, a single agent for work inside Google Workspace that Google says runs on Gemini and Anthropic Claude models. Reuters (opens in a new tab)
Official uploads from the organisations' own YouTube channels. Nothing loads from YouTube until you press play, then it uses youtube-nocookie.com.
Prices, quality, speed and context for 109 models. Use the model finder. Latest launches are in AI news.
Vendor-neutral. Listing an agent is not a recommendation. Checked 1 Nov 2026.
What changed this month for Grok Bot, ChatGPT Work, Gemini, Claude and Microsoft 365 Copilot, each from the maker's own page, is on AI news: AI bots and agents.
6 leaders and researchers who shape the models, chips and data centres. Roles are as stated on each linked page. Photos are freely licensed and credited on each card.
Chips 
NVIDIA
Co-founder, president and CEO of NVIDIA.
Source: Wikipedia: Jensen Huang (opens in a new tab), accessed 1 November 2026
Hyperscalers
Microsoft
Chairman and CEO of Microsoft, CEO since 2014.
Source: Wikipedia: Satya Nadella (opens in a new tab), accessed 1 November 2026
Hyperscalers
Alphabet
CEO of Google since 2015 and of Alphabet since 2019.
Source: Wikipedia: Sundar Pichai (opens in a new tab), accessed 1 November 2026
Hyperscalers 
Microsoft AI
CEO of Microsoft AI. Co-founded DeepMind and Inflection AI.
Source: Wikipedia: Mustafa Suleyman (opens in a new tab), accessed 1 November 2026
Labs
xAI, Tesla, SpaceX
CEO of Tesla and SpaceX.
Source: Wikipedia: Elon Musk (opens in a new tab), accessed 1 November 2026
Researchers
Stanford University
Professor of computer science at Stanford. Created ImageNet, the dataset behind the 2010s advances in computer vision.
Source: Wikipedia: Fei-Fei Li (opens in a new tab), accessed 1 November 2026
Not fact. Everything in this section is a rumour, a theory or a forecast. Each card says where it circulates and sits beside what is actually known, with its own source. We do not endorse any of it, and none of it is investment advice.
Rumour (0)Speculation (0)Prediction (3)
The claim
Joachim Klement, head of market strategy at Panmure Liberum, says his core conviction is that the AI bubble bursts in 2027 or 2028. His year-end 2027 target for the S&P 500 is 5,000 points, which he put at 36% below the level at the time.
Circulates in: Bloomberg, via Yahoo Finance: AI Bubble Risks Worst S&P 500 Crash Since 2008, Strategist Says (opens in a new tab)
What is known
The claim
Capital Economics forecasts the S&P 500 at 6,500 by the end of 2027, about 16% lower, if air keeps leaking out of the AI trade or the bubble bursts. It also expects a final tech-led rally in 2026 first.
Circulates in: CNBC: Firm that correctly called the AI rally slowdown says here's where to hide out (opens in a new tab)
What is known
The claim
Gartner analyst Will Sommer predicts some US$3.5tn of AI infrastructure spending could be written off by the end of 2030, and that 80% of neoclouds will stop operating independently by 2030, bought or gone under.
Circulates in: The Telegraph: The tech 'death spiral' that could destroy the AI boom (opens in a new tab)
What is known
Checked 1 November 2026. Rumour means an unconfirmed report of a specific event. Speculation means a theory or interpretation. Prediction means a forecast with a date or a number. See all rumours, theories and predictions Not investment advice.
13 coding agents with what each maker publishes about free use and the lowest paid plan. Listed in our usual order, not ranked.
| Agent | Maker | Free plan or free use | From, a month | MCP | Open source | Source |
|---|---|---|---|---|---|---|
| Claude Code | Anthropic | No | US$20 | Yes | No | claude.com (opens in a new tab) |
| OpenAI Codex | OpenAI | Yes | US$8 | Yes | Yes | developers.openai.com (opens in a new tab) |
| GitHub Copilot (agent mode and coding agent) | GitHub (Microsoft) | Yes | US$10 | Yes | No | github.com (opens in a new tab) |
| Cursor | Anysphere | Yes | US$8 | Yes | No | cursor.com (opens in a new tab) |
| Devin and Windsurf | Cognition | Yes | US$20 | – | No | devin.ai (opens in a new tab) |
| Replit Agent | Replit | Yes | US$20 | – | No | replit.com (opens in a new tab) |
| Lovable | Lovable | Yes | – | Yes | No | lovable.dev (opens in a new tab) |
| Kiro | Amazon (AWS) | Yes | US$20 | – | No | kiro.dev (opens in a new tab) |
| Google Antigravity | Yes | Free | – | No | antigravity.google (opens in a new tab) | |
| Grok Build | xAI | No | US$8 | Yes | No | x.ai (opens in a new tab) |
| Muse Code | Meta | No | – | – | No | dev.meta.ai (opens in a new tab) |
| OpenHands | All Hands AI | Yes | US$500 | Yes | Yes | all-hands.dev (opens in a new tab) |
| Cline | Cline | Yes | – | Yes | Yes | cline.bot (opens in a new tab) |
A dash means the maker does not publish it. Prices are list prices in US dollars, see each agent page for plan names and limits.
The 42 agents and frameworks on this page, grouped by what they do. Open any one for its sourced profile.
Claude Code, OpenAI Codex, GitHub Copilot (agent mode and coding agent), Cursor, Devin and Windsurf, Replit Agent, Lovable, Kiro, Google Antigravity, Grok Build, Muse Code, OpenHands, Cline
Claude in Chrome, Claude computer use tool, Browser Use, Manus, Perplexity Comet
Grok Bot, ChatGPT agent mode, Claude (apps, Cowork and research), Gemini app agent mode, Microsoft 365 Copilot
Microsoft Copilot Studio, Salesforce Agentforce, Amazon Bedrock AgentCore, Gemini managed agents (Antigravity agent), Claude Managed Agents, Zapier Agents, n8n
OpenAI Agents SDK, Claude Agent SDK, Agent Development Kit (ADK), LangGraph and LangSmith, AutoGen, CrewAI, Mastra
Gemini Deep Research agent (API), Perplexity (Pro and Max research)
An AI that can carry out tasks for you, like booking, filling in forms or writing and running code, rather than only answering questions.
Why it matters A chat assistant tells you how to do something. An agent can go and do it, and should ask you before anything important.
A chatbot answers in a chat window. An agent can use apps, websites and files to get a job done, step by step.
They can make mistakes, so good agents ask for your approval before they send, buy or delete anything. Start with small, low-risk jobs.
Yes. An agent may look at a screen many times to finish one job, and each look uses computing.
Why it matters One screenshot can use around 1,000 to 1,800 input tokens, so long agent runs add up fast.
Source: Anthropic computer use documentation
Use the Agent Finder: pick the job you want done and your skill level, and it shows agents that fit, each with links to the maker's own pages.
There is no single best one. The main coding agents are Claude Code, OpenAI Codex, GitHub Copilot, Cursor, Devin and Windsurf, Replit Agent, Kiro and Google Antigravity. The coding agents table on this page shows each one's free plan, starting price and source.
Why it matters Pick by where you work: in a terminal, in your editor or in a browser.
Source: Each maker's pricing page, checked 1 November 2026
Yes. GitHub Copilot, Cursor, OpenAI Codex, Google Antigravity and Perplexity Comet have free plans or free use, and frameworks such as the OpenAI Agents SDK, Claude Agent SDK, Google ADK and Mastra are free and open source. You still pay for model use where the agent calls a paid API.
Why it matters Free plans usually have tight usage limits, so check the linked pricing page.
Source: Each maker's pricing page or repository, checked 1 November 2026
