May 2026 is the month AI agents stopped being a promise and became something you can actually buy and deploy. Microsoft Agent 365 hit general availability on May 1. Claude Cowork has been shipping on macOS and Windows since early 2026. OpenAI's Operator drives a real browser for you. Cursor 3 runs fleets of parallel coding agents. Perplexity Computer coordinates 19 backend models for a single task.
Gartner predicts 40% of enterprise applications will include task-specific AI agents by the end of 2026. If you are trying to figure out which one is real and which one is marketing, this guide is for you.
All pricing sourced from official websites as of May 2026.
Sources: anthropic.com/pricing, openai.com/chatgpt/pricing, cursor.com/pricing, perplexity.ai, microsoft.com/en-us/microsoft-365/copilot
Quick Comparison Table
| Platform | Entry Price | On Your Desktop? | Browser Access? | File System? | Best For |
|---|---|---|---|---|---|
| Claude Cowork | $20/mo (Pro) | ✅ macOS + Windows | ✅ | ✅ Full | Document creation, file management |
| Microsoft Agent 365 | $30/user/mo (Copilot) | ❌ Cloud only | Limited | ❌ | M365 enterprise workflows |
| ChatGPT Operator | $200/mo (Pro) | ❌ Cloud only | ✅ Full browser | ❌ | Web tasks, reservations |
| Cursor 3 Agents | $20/mo (Pro) | ✅ IDE only | ❌ | ✅ Codebase | AI-native coding, parallel agents |
| Google Gemini | $19.99/mo (Pro) | ❌ Cloud only | ✅ | ❌ | Google Workspace, long documents |
| Perplexity Computer | $200/mo (Max) | ✅ Desktop app | ✅ Full browser | ❌ | Multi-source research |
Claude Cowork (Anthropic) — Best Desktop Agent for Document Work
Pricing: Free (Sonnet 4.5, file creation) → Pro $20/mo (all models, Cowork access) → Max $100/mo (5x usage) → Max $200/mo (20x usage)
Claude Cowork is a desktop agent that runs on macOS and Windows. Unlike cloud-only assistants, it can directly interact with your computer — creating and editing files, organizing folders, synthesizing research into documents, and running multi-step tasks autonomously.
Strengths:
- Full desktop access: reads and writes local files, opens apps, works while you do other things
- Document creation: native Word, Excel, PowerPoint, and PDF output you can edit immediately
- App connectors: Slack, Gmail, Google Drive, Notion, Figma, and more
- Scheduled tasks: set up recurring agent runs
- Strongest model for agentic tasks: Claude Opus 4.7 topped Palisade Research's self-replication benchmark at 81%, far ahead of GPT-5.4 at 33%
Weaknesses:
- Heavy Cowork use burns through token allocation fast
- Desktop app only — no mobile or web-only Cowork
- True autonomous mode needs at least the $100/mo Max plan
Best for: Document-heavy workflows, research synthesis, file organization. Anyone who needs an AI that works with local files rather than just chat.
Microsoft Agent 365 / Copilot Cowork — Best for M365 Enterprise
Pricing: Copilot Cowork is included in Microsoft 365 Copilot ($30/user/mo add-on). Agent 365 requires the new E7 bundle ($99/user/mo — includes E5 + Copilot + Entra Suite).
Microsoft launched Agent 365 on May 1, 2026. It is a different product from Copilot Cowork. Copilot Cowork runs task-completion workflows inside M365 apps (Outlook, Teams, Word, Excel, PowerPoint). Agent 365 is an enterprise agent-runtime layer that supports multiple agents per user with Entra-managed identities.
Strengths:
- Deepest M365 integration: works inside Outlook, Teams, Excel, Word, PowerPoint natively
- Multi-agent governance: Agent 365 supports multiple agents per user with Entra identity management
- Enterprise security: EU Data Boundary, SOC 2, GDPR
- Work IQ: grounded in your emails, files, meetings, and calendar
Weaknesses:
- Requires existing M365 subscription — the $30/mo is an add-on, not standalone
- Copilot Cowork is still in limited availability (Frontier program)
- Agent 365 requires E7 bundle at $99/user/mo — steep for small teams
- Uses Anthropic models as a subprocessor inside Microsoft's Cowork
Best for: Organizations already on Microsoft 365. Heavy Outlook/Teams/SharePoint users. Compliance-heavy industries (healthcare, finance).
Source: microsoft.com/en-us/microsoft-365/copilot
ChatGPT + Operator (OpenAI) — Best Web Agent
Pricing: Free (GPT-4o, daily limits) → Plus $20/mo (GPT-5.4, Custom GPTs) → Pro $200/mo (unlimited, Operator, Deep Research)
ChatGPT's agent capabilities come in two forms: Custom GPTs (which can call external APIs) and Operator (a browser-driving agent available on the Pro tier). Operator can autonomously navigate websites — ordering food, comparing prices, filling out forms.
Strengths:
- Largest ecosystem: most third-party GPTs, most integrations
- Operator drives a real browser: can complete multi-step web tasks like booking travel
- Deep Research: spends 5-30 minutes on multi-source investigations with cited reports
- GPT-5.4 Thinking: reasoning integrated into the main model, no separate product
Weaknesses:
- No local file system access — cannot read or write files on your computer
- Operator is Pro-only at $200/mo and still US-only as of May 2026
- Deep Research query limits reduced: reports of limits dropping from ~500/day to as low as 20/month
Best for: General-purpose AI chat, web-based tasks that require browser interaction, users who want the largest ecosystem of third-party tools.
Source: openai.com/chatgpt/pricing
Cursor 3 Agents — Best AI-Native IDE
Pricing: Free (2,000 completions) → Pro $20/mo ($20 compute included) → Ultra $200/mo (higher compute)
Cursor 3 shipped April 2, 2026, and is the biggest release since Cursor forked VS Code. The headline feature is the Agents Window — a dedicated panel that lets you run multiple AI coding agents in parallel across local machines, worktrees, SSH, and cloud environments.
Strengths:
- Parallel agents: run multiple agents simultaneously, each in its own context
- Background Agents: work in isolated VMs on their own Git branches, open PRs when finished
- Cloud Agents: can be triggered from Slack, GitHub, or mobile
- Composer 2 (in-house model): more cost-efficient than routing every request to frontier models
- Design Mode: annotate UI elements in the browser to give the agent precise targets
Weaknesses:
- Compute-based pricing: using premium models burns through the $20 allocation fast
- IDE-only: Cursor is a code editor, not a general-purpose agent
- Overage charges at $0.04 per additional agent request
Best for: Software developers who want AI agents that write, refactor, and deploy code autonomously.
Google Gemini — Best for Google Ecosystem + 1M Context
Pricing: Free (basic Gemini) → AI Pro $19.99/mo (Gemini 3.1 Pro) → AI Ultra $125/mo (maximum usage)
Gemini's agentic strengths are tightly coupled to Google Workspace. It can draft emails, create documents, analyze spreadsheets, and summarize meetings directly within Google's apps. The 1M token context window is its technical standout — upload an entire codebase or a 1,500-page document and ask questions.
Strengths:
- Deepest Google integration: Gmail, Docs, Sheets, Slides, Drive, Calendar, Maps
- 1M token context window: unmatched for long-document analysis
- Gemini 3.1 Pro: 94.3% on GPQA Diamond, led both GPT-5.2 and Claude Opus 4.6 at the time
- Gemini Spark (announced at I/O 2026): closing the agentic gap with desktop capabilities
Weaknesses:
- Less agentic autonomy: cannot take over your desktop or browse websites independently
- Strengths are reactive: answering questions and drafting within Google apps, not proactive multi-step execution
- Outside Google ecosystem: fewer integrations than competitors
Best for: Users embedded in Google Workspace. Long-document analysis. Android users.
Source: blog.google/products/gemini
Perplexity Computer — Best Research Agent
Pricing: Free (basic search) → Pro $20/mo (Pro Search, Deep Research) → Max $200/mo (Computer mode, 10,000 credits)
Perplexity's Computer mode (launched February 25, 2026) is a general-purpose digital worker that accepts high-level objectives, decomposes them into subtasks, and delegates to whichever of its 19 backend models is best suited. It is the most innovative approach to agent orchestration currently shipping.
Strengths:
- Best-in-class research: answers with cited sources you can verify
- Computer mode: coordinates 19 models for complex analytical workflows
- Deep Research: produces detailed reports with inline citations
- Free tier is genuinely useful: unlimited basic searches
Weaknesses:
- Research-focused: not designed for file management, document creation, or business workflow automation
- Computer mode is $200/mo Max tier only
- Deep Research query limits reduced for some Pro subscribers
Best for: Anyone who needs verifiable research, competitive analysis, or cited reports. Not a general-purpose productivity agent.
Feature Comparison Matrix
| Feature | Claude Cowork | MS Agent 365 | ChatGPT Operator | Cursor 3 | Gemini | Perplexity Computer |
|---|---|---|---|---|---|---|
| Local file system | ✅ Full | ❌ | ❌ | ✅ Codebase | ❌ | ❌ |
| Desktop app | ✅ macOS/Win | ❌ | ❌ | ✅ IDE | ❌ | ✅ Desktop |
| Browser autonomy | ✅ | Limited | ✅ Full | ❌ | ❌ | ✅ Full |
| Multi-step planning | ✅ | ✅ | ✅ | ✅ | Partial | ✅ |
| Scheduled tasks | ✅ | ✅ | ❌ | ✅ | ❌ | ❌ |
| Max context | 200K tokens | N/A | 1M tokens | Varies | 1M tokens | N/A |
| Model orchestration | Single | Anthropic backend | Single | Multi-model | Single | 19 models |
| Enterprise SSO | ✅ Team/Ent | ✅ Entra | ✅ Enterprise | ❌ | ✅ Workspace | ❌ |
| Free tier | ✅ Sonnet 4.5 | ❌ | ✅ GPT-4o | ✅ Limited | ✅ Basic | ✅ Basic |
Pricing Breakdown by Platform
| Tier | Claude | MS Agent 365 | ChatGPT | Cursor | Gemini | Perplexity |
|---|---|---|---|---|---|---|
| Free | Sonnet 4.5, docs | ❌ | GPT-4o, daily caps | 2K completions | Basic chat, Android | Basic search |
| $10-20/mo | Pro $20 (Cowork!) | ❌ | Plus $20 | Pro $20 | Pro $19.99 | Pro $20 |
| $30-50/mo | ❌ | Copilot $30/user | ❌ | ❌ | ❌ | ❌ |
| $100-125/mo | Max 5x $100 | ❌ | ❌ | ❌ | Ultra $125 | ❌ |
| $200/mo | Max 20x $200 | ❌ | Pro $200 (Operator!) | Ultra $200 | ❌ | Max $200 (Computer!) |
| Enterprise | Team/Ent quoted | E7 $99/user | Enterprise quoted | ❌ | Enterprise quoted | ❌ |
Quick Decision Guide
| If you need... | Choose | Entry price |
|---|---|---|
| Desktop agent that edits your files | Claude Cowork (Pro) | $20/mo |
| Enterprise agent on Microsoft stack | MS Agent 365 (Copilot) | $30/user/mo |
| Browser-based web tasks | ChatGPT Operator (Pro) | $200/mo |
| AI-native coding with parallel agents | Cursor 3 (Pro) | $20/mo |
| Google Workspace integration | Gemini (Pro) | $19.99/mo |
| Research with verifiable citations | Perplexity (Pro) | $20/mo |
Summary
The AI agent market bifurcated in two directions in 2026. Desktop agents (Claude Cowork, Cursor) work on your machine with your files. Cloud agents (Microsoft Agent 365, ChatGPT Operator, Gemini, Perplexity) run in the vendor's infrastructure and work with your SaaS apps.
For most individuals and small teams, Claude Cowork at $20/mo is the best value — it is the only major agent platform that combines desktop file access with cloud app connectors and document creation in a single subscription.
For M365-centric enterprises, Agent 365 is the obvious choice, but the $99/user/mo E7 bundle is a significant investment — start with Copilot Cowork at $30/user/mo and upgrade once your team has built agentic workflow muscle.
For developers shipping code, Cursor 3 at $20/mo with its parallel agent architecture is the most productive option, but watch the compute-based pricing if you use premium models heavily.
Pricing sourced from official websites as of May 2026. Check each platform's pricing page for the most current rates.
Try Claude Free
Anthropic's AI assistant with Cowork desktop agent. Start free.
Get Started — from $20/mo