AgentsPick an agent for research and the web
Do you need cited answers, a finished report, or an agent that clicks through websites for you? Every lab now ships a deep-research mode and most ship a browser agent, so start with the one you already pay for.
Deep research and browser agents from the frontier labs
ChatGPT Work, Codex & Agents SDK
Choose it if: You pay for ChatGPT: deep research writes cited reports, ChatGPT Work finishes multi-step web tasks in its own cloud browser, and the ChatGPT Atlas browser brings ChatGPT into your own browser.
The cloud browser in ChatGPT Work is on paid plans (not Free or Go). Codex is included in ChatGPT Free, Go, Plus, Pro, Business and Enterprise, with usage that grows by tier.
Gemini, Jules & Gemini Enterprise
Choose it if: You use Google: Gemini Deep Research writes cited reports, and Gemini in Chrome auto browse (AI Pro or Ultra, US and India) fills forms and shops across sites.
Google AI Pro $19.99/mo; Ultra $99.99/mo (5x Pro usage) or $199.99/mo (20x). Gemini Enterprise Business $14/user/mo, Standard from $21/user/mo.
Claude, Claude Code & Claude Agent SDK
Choose it if: You pay for Claude: Research mode does cited multi-step research, and Claude in Chrome (all paid plans) works across tabs in your own signed-in browser.
Pro $20/mo ($17/mo billed annually), Max from $100/mo (5x) or $200/mo (20x). Claude Code is included in paid plans and shares the plan's usage limits. Max and Team plans include monthly API credits that cover the Agent SDK.
Grok & Grok Bot
Choose it if: You use Grok or X: DeepSearch digs through many sources and writes a concise report, and Grok Bot agents keep working on their own cloud computer.
The X Subscription: Free; Plus $8/mo; Premium $30/mo; Super $100/mo; Ultra $200/mo (annual discounts available). Plus doesn't include Grok Bot.
Microsoft 365 Copilot
Choose it if: Your company runs on Microsoft 365: Researcher does cited research grounded in your company files and the web.
Microsoft 365 Copilot $30/user/mo (annual) on top of a qualifying Microsoft 365 plan. Long-running agentic features use usage-based billing.
Research and browser specialists
Perplexity & Comet
Choose it if: You want fast answers with citations, and Comet, a browser that can act on pages.
Check perplexity.ai for current plans. We couldn't verify today's tier prices on an official page.
Manus
Choose it if: You want long research reports, slide decks and scheduled research tasks.
Free (300 daily credits, Lite model); Pro from $20/mo or $40/mo; Team from $20/seat/mo. Annual billing saves 17%.
Open source, on your own machine
OpenClaw
Choose it if: You want a free, open-source agent you message from chat apps, able to browse with a local model.
Free and open source (MIT). There is no subscription or paid edition; you pay only for any cloud model you connect. Local models through Ollama cost nothing to run.
Hermes Agent & Hermes Desktop
Choose it if: You want a free, open-source desktop agent with a built-in Local Models screen.
Free and open source (MIT). Local models cost nothing to run. Cloud models are optional and billed by the provider you choose (Nous Portal is an optional subscription).
More research and browser agents
Windows 365 for Agents
Choose it if: IT teams that need agents to use Windows apps under company policy
Pay as you go: $0.40 per Cloud PC hour in the US, $0.52 in Europe and the UK, and $0.80 in South America. An always-available Cloud PC adds $5/month in the US. Billed through Azure, with 50 free hours to start through Copilot Studio. Autopilot uses usage-based billing with no separate published price.
Amazon Bedrock AgentCore & Nova Act
Choose it if: Teams already building on AWS
AgentCore Browser and Code Interpreter: $0.0895 per vCPU-hour and $0.00945 per GB-hour, billed per second, and idle time during I/O waits is free. Nova Act: $4.75 per agent hour.
Browserbase
Choose it if: Developers building browser agents
Free (1 browser hour); Developer $20/mo (100 hours, then $0.12/hour); Startup $99/mo (500 hours, then $0.10/hour); Scale custom.
Local and private option
For private document Q&A, see our use case on running a local model over your own files, and the local agents guide (/learn/local-agents) for OpenClaw and Hermes Agent. Open the model picker →
Guardrails
- Check the citations. Agents still get things wrong.
- Don't let a browser agent enter payment details without confirming.
- Browser agents can be misled by hidden text on web pages. Start on sites you know.