AI Agent Comparison: Manus, Flowith, Suna, Convergence AI, Perplexity Labs – Pros & Cons Summarized
The Best Commander2025. 6. 2. 05:09
In today’s rapidly evolving AI landscape, AI agents have emerged as indispensable tools for automating complex, multi-step tasks that previously required human oversight. From web scraping and data analysis to content generation and report creation, these intelligent agents promise to revolutionize productivity across industries. Whether you’re a developer aiming to build a sophisticated application, a marketer who needs automated research, or a business leader looking for cost-effective solutions, choosing the right AI agent can make or break your workflow.
This post presents a comprehensive, head-to-head comparison of five prominent AI agents in 2025:
Manus
Flowith
Suna
Convergence AI (Proxy)
Perplexity Labs
Throughout this analysis, we rely on trustworthy sources, including official documentation, reputable news outlets, and expert reviews to ensure factual accuracy. If you’re ready to discover which AI agent is truly worth your time and budget in 2025, let’s jump right in.
1. What Are AI Agents, and Why Do They Matter?
1.1 Definition & Context
An AI agent is a software program that autonomously carries out tasks on behalf of a user, often by planning, reasoning, and taking actions with little to no continuous human intervention. Unlike simple chatbots that generate text responses in a conversational interface, AI agents can:
Navigate websites: Fill out forms, click buttons, bypass login screens (subject to CAPTCHA restrictions).
Execute code: Run scripts to analyze data, generate reports, or create dashboards.
Collaborate with tools: Integrate with IDEs, data analytics platforms, or no-code automation software.
Maintain memory: Store and recall past interactions to personalize future workflows.
The importance of AI agents has surged as organizations face ever-increasing volumes of data and demand faster turnaround times. By delegating repetitive or complex processes to intelligent agents, teams can focus on high-value, strategic work—ultimately boosting productivity, efficiency, and scalability.
1.2 Key Use Cases
Business Intelligence & Market Research: Automatic aggregation of data from multiple sources (e.g., financial websites, competitor portals), followed by synthesis into dashboards or reports.
Content Creation & SEO: Generating blog posts, social media drafts, or video scripts by blending real-time web search with language generation.
Web Automation & E-commerce: Monitoring price fluctuations, updating product listings, placing orders, or scraping customer reviews.
Software Development Assistance: Writing boilerplate code, debugging, generating unit tests, and automating CI/CD pipelines.
Academic & Scientific Research: Collecting scholarly articles, summarizing findings, creating annotated bibliographies, and building interactive visualizations.
2. Overview of the Five AI Agents
Below is a quick snapshot of the six AI agents we will examine in detail:
Manus Developer: Monica (Butterfly Effect AI), Singapore Launch Date: March 6, 2025 Target Use Case: Fully autonomous, multimodal agent for complex, end-to-end workflows (e.g., web dev, stock analysis).
Flowith Developer: Flowith (Hangzhou, China) Launch Status: Closed beta (invite-only) in 2025 Target Use Case: Ultra-long-context (up to 10M tokens), multimodal projects (3D games, video generation, long-form writing).
Suna Developer: Kortix AI Corp, United States (open-source) Launch Status: Publicly available cloud version + GitHub repo Target Use Case: Versatile, open-source generalist agent—ideal for research, data scraping, data analysis, file management.
Convergence AI (Proxy) Developer: Convergence AI, London, UK Launch Status: Proxy 1.0 launched late 2024 (public beta) Target Use Case: Browser-based automation (web task delegation, data entry, email handling) with easy-to-use “Template Studio.”
Perplexity Labs Developer: Perplexity AI, Inc., USA Launch Status: Labs feature released May 2025 (requires Pro subscription) Target Use Case: Research-centric agent—combines real-time web search, code execution, and data visualization to produce reports, dashboards, and mini web apps.
Each of these platforms has a unique positioning, whether it’s emphasizing open-source transparency (Suna), massive context windows (Flowith), or enterprise-grade security (Manus). The next sections will unpack their offerings in depth.
3. Detailed Breakdown: Manus
3.1 Core Features
Fully Autonomous Task Execution Manus is designed to “bridge minds and hands,” meaning it not only thinks but also takes direct actions—whether that’s writing and deploying code, performing data analysis, or interacting with web services.
Multimodal Capabilities Although primarily text-driven, Manus can integrate with external tools (e.g., code editors, browsers) to handle tasks such as:
Building and deploying simple web applications.
Scraping financial data to generate stock correlation reports.
Conducting travel planning with itinerary generation.
Asynchronous Processing & Memory
Cloud-based GPU Execution: Tasks run asynchronously in the cloud, freeing local resources.
Persistent Memory: Manus stores conversation histories and preferences to personalize output over time—like prioritizing specific data sources or summarization styles.
Visible Workflow Transparency Manus displays its step-by-step plan in a side panel. Users can see intermediate actions—e.g., “Navigating to Yahoo Finance → Extracting S&P 500 data → Performing correlation analysis.” This transparency appeals to non-technical users who wish to understand, audit, or intervene in the process.
3.2 Pros
High Autonomy: Handles end-to-end workflows with minimal human intervention, making it best for users who need “set it and forget it” solutions.
Rich Tool Integration: Out-of-the-box connectors to common development and data analysis tools.
Competitive Pricing (Beta Stage)
Starter Plan: \$39/month per seat (minimum 5 seats = \$195/month) — 3,900 credits, 2 concurrent tasks (credits are consumed per operation; major tasks can run 30 minutes to over an hour).
Pro Plan: \$199/month per seat — 20,000 credits, up to 5 concurrent tasks with priority access during peak times.
Global Recognition & Support: Official backing from China’s AI initiatives, partnerships with Alibaba’s Qwen, and rapid user adoption (2 million waitlisted users).
3.3 Cons
Invite-Only Beta: Access is limited to invite-code holders; resale value of codes can exceed \$1,000.
Stability Issues: Under heavy demand, users have reported server crashes and higher failure rates compared to competitors, especially when dealing with paywall-protected websites or complex CAPTCHAs.
Learning Curve: Configuration and workflow design can be daunting for first-time users unfamiliar with agent frameworks.
Cost Concerns: Although \$39/month is competitive, per-task costs can accumulate—complex projects may consume hundreds of credits quickly (e.g., a 10-minute NBA efficiency chart task can cost ~200 credits).
4. Detailed Breakdown: Flowith
4.1 Core Features
Ultra-Long Context Window (up to 10M Tokens) Flowith’s flagship Agent Neo can maintain massive conversations without truncation, enabling:
Entire books or research papers to be loaded and processed in a single workflow.
Long-form writing projects (e.g., drafting a novel with coherent character arcs).
Complex 3D game development prompts that span months of iterative design.
Multimodal Support (Text, Images, Video, 3D)
Text Models: GPT-4.1-mini, Claude 3.5 Haiku, Gemini 2.5 Flash, Grok 3-mini.
Image Models: DALL·E 3, Flux 1.1 Pro, Recraft V3.
Video Models: Veo 2, Kling V2.
3D Generation: Experimental support for generating simple 3D assets (e.g., meshes, textures) within workflows.
Knowledge Gardens (Data Marketplace)
Data Sharing: Users can create, sell, or utilize curated data sets to train or fine-tune specialized agents.
Collaborative Learning: Access to community-generated (or verified) data sets for industry-specific tasks (e.g., medical literature, legal precedents, geospatial data).
Flow Mode & Oracle Mode
Flow Mode: Standard AI canvas—drag-and-drop nodes representing tasks, conditional branches, and loops. Ideal for rapid prototyping of pipelines.
Oracle Mode: A specialized mode for long-duration, multi-phase projects—tasks can run for days, weeks, or even months, storing intermediate states and logs for later resumption.
4.2 Pros
Unmatched Context Capacity: Perfect for researchers or creators handling massive texts (e.g., legal contracts, encyclopedias) without losing coherence.
Highly Customizable: Agents can be manually tuned using specialized data sets from Knowledge Gardens.
Team Collaboration:
Team Tiers: From Free (1,000 one-time credits) to Professional (\$13.93/month, 20,000 credits) to Ultimate (\$29.49/month, 50,000 credits) to Infinite (\$249.95/month, 500,000 credits).
Early Adopter Perks: 30–50% off on yearly plans; priority upload capacities (400K tokens/upload or 100M tokens storage).
Multimodal Prowess: Allows one agent to mix text, image, and video tasks in the same workflow—ideal for complex marketing campaigns (e.g., automatically generate a script → storyboard frames → draft a promotional video).
4.3 Cons
Invite-Only Access: Still closed beta—requires an invitation code, raising barriers to entry.
Opaque Team Info: Limited public details on the founding team, dev roadmap, or security audits, making some users question long-term reliability.
Mobile Experience Weakness: Reported lag and unreliable syncing between web and mobile apps.
Steep Learning Curve: The 2D canvas interface, while powerful, can overwhelm newcomers expecting a simple chat-like experience.
4.4 Flowith Pricing Summary
Plan
Monthly (Yearly Billing)
Credits (Monthly)
Context Storage
Features
Starter (Free)
$0
1,000 (one-time)
N/A
Basic AI models, Agent Neo access, community support
Professional (Pro)
$13.93 (30% off)
20,000
10M tokens Knowledge Garden
Advanced AI models, Video & Ultra AI (30 uses/day), Team collab
Ultimate
$29.49 (40% off)
50,000
100M tokens Knowledge Garden
Everything in Pro + Ultra AI (70 uses/day), priority support
Infinite
$249.95 (50% off)
500,000
Unlimited
Enterprise features, dedicated support, advanced team collab
Enterprise
Custom
Unlimited
Custom
Unlimited model usage, custom storage & credits, VIP support
Table 1: Flowith Pricing & Feature Comparison
5. Detailed Breakdown: Suna
Note: Do not confuse Suna (this AI agent) with Suno (the audio/music generation platform). Suna is developed by Kortix AI Corp and is fully open-source.
5.1 Core Features
Open-Source Generalist Agent GitHub Repository License: MIT (100% source code available). Ideal for developers or organizations that require full transparency and auditability.
Browser Automation & Web Scraping
Built-in modules to navigate websites, handle logins (limited by CAPTCHAs), extract data, and save outputs (CSV, JSON, PDF).
Users can specify scraping rules via CSS selectors or XPath.
Data Analysis & File Management
Integration with pandas: Users can import CSVs, run data transformations, generate dataframes, and export processed files.
Excel/Spreadsheet Automation: Generate and populate XLSX files with analysis results.
Report Generation: Create PDF or Markdown summaries based on scraped or processed data.
Local & Cloud Deployment
Cloud Version (suna.so): Simple signup → Google login → start chatting. All infrastructure is managed by Kortix AI.
Self-Hosted Option: Clone the GitHub repo, install dependencies (Supabase, Redis, OpenAI/Anthropic API keys), and run locally—ensuring full data sovereignty.
5.2 Pros
100% Open Source: Eliminates vendor lock-in, allows full customization (add custom modules, integrate proprietary systems). Ideal for privacy-conscious organizations.
Flexible Deployment:
Cloud (suna.so): Free tier + Pro + Custom. No infrastructure overhead for developers who want a quick start.
On-Premises: Comply with internal security/audit requirements (medical, legal, government sectors).
Cost-Effective Pricing:
Free Plan: \$0—60 minutes/month on public projects, limited to basic model (no private projects).
Pro Plan: \$20/month—120 minutes/month, private projects, full Suna model.
Custom Plan: \$50/month—360 minutes/month (6 hours), ideal for power users or small teams.
Rapid Onboarding:
Cloud Version: Start within minutes—no coding required.
Developer-Friendly: Clone repo, follow README instructions, and run locally.
5.3 Cons
Limited Usage Minutes:
Free (1 hour/month) and Pro (2 hours/month) may not suffice for heavy users doing long scrapes or large data analyses.
Users must purchase custom plans or purchase additional minutes as needed.
Lack of Advanced Multimodality:
No built-in support for video generation, 3D, or large context windows.
Focuses on text-based research and data tasks.
Relatively Young Project:
Active community is still growing—documentation and community-contributed modules are fewer compared to older, closed-source platforms.
Interface Is Basic:
Chat UI with CLI-like interactions—no drag-and-drop canvas or visual workflow designer.
5.4 Suna Pricing Summary
Plan
Price (Monthly)
Usage Minutes
Features
Free
$0
60 minutes
Public projects, basic model (limited capabilities)
Pro
$20
120 minutes
Private projects, full Suna model (browser + API)
Custom
$50
360 minutes
Everything in Pro, customizable usage (6 hours)
Table 2: Suna Pricing Summary
6. Detailed Breakdown: Convergence AI (Proxy)
6.1 Core Features
Browser-Based Agent (No Code Required) Proxy operates entirely in a web browser. Users can create automations by typing natural language prompts. Under the hood, Proxy’s agent simulates a human browsing experience—clicking buttons, filling forms, and navigating pages.
Template Studio & Community Hub
Template Studio: Pre-built automation templates (e.g., “Price Tracker,” “Auto Email Responder”) that users can clone and customize.
Template Hub: A community-driven library where users share their automations—newcomers can start by importing a proven template, reducing trial-and-error.
Parallel Sessions & Automation Limits
Free Plan: 5 sessions/day, 1 automation.
Pro Plan ($20/month): Unlimited sessions, up to 20 concurrent automations, access to 20 saved automations.
Real-Time Process Visualization Proxy displays its web interactions in real time. Users can watch how it logs in to Gmail, extracts contacts, or fills out a product order—then edit or pause at any step.
6.2 Pros
Ease of Use:
Web-based interface requires zero installation—ideal for non-technical teams.
Natural language prompts make automation accessible to anyone familiar with typing instructions.
Deep Research (Tier available in free & Pro): Aggregates real-time search results with source citations—ideal for quick research.
Labs Mode (Requires Pro $20/month or Enterprise):
Runs longer, multi-step workflows (≥10 minutes total processing) autonomously.
Can execute code (Python) in a sandboxed environment to analyze data, generate charts, or create spreadsheets.
Create Mini Web Apps: With simple configuration, generate an interactive dashboard (e.g., “Global AI Funding Tracker”) that updates automatically.
Multiple Model Support
Sonar Pro (High Context Reasoning), Sonar (Lightweight), Sonar Deep Research (Exhaustive analysis), r1-1776 (Offline model).
Access to GPT-4, Claude 3, Mistral Large, and Llama 3 for Pro users.
Spreadsheet & Dashboard Generation
Perplexity Labs can write Python code to manipulate data (e.g., load CSV → filter rows → compute pivot tables) and then output a Google Sheet or Excel file.
Alternatively, it can create interactive charts using Matplotlib, embed them in a dashboard, and host a public link.
Extensive Citation & Source Transparency All research-based answers display source citations at the end of paragraphs (e.g., ). This level of transparency is crucial for academic or professional research, ensuring verifiability.
7.2 Pros
Holistic Research & Analysis:
Combines real-time web search with AI-driven code execution → single-step process to build reports.
Eliminates the need to switch between browser search, coding environment, and spreadsheet.
Pro-Level Models at $20/Month:
Pro Subscription: $20/month (or $200/year) unlocks Labs, GPT-4, Claude 3, Mistral, Llama, and up to 10 minutes of autonomous runs per project.
Enterprise Pro: $40/month per seat (or $400/year) for organizations (<250 employees) with enhanced security & data retention.
User-Friendly Workflow:
Single Interface: Choose “Labs” mode, type prompt (e.g., “Gather latest AI investment data, analyze, and build a dashboard”), and let the agent handle the rest.
Auto-Generated Assets: Charts, spreadsheets, and web apps are organized under the user’s project tab—easy to revisit or share.
7.3 Cons
Labs Access Requires Pro:
Free users can only access basic search and limited Deep Research → no code execution or dashboard creation.
Token-Based Pricing for API:
Integrating Labs into external apps (e.g., building a custom internal dashboard) incurs token costs:
Input Tokens: $2–$3 per million
Output Tokens: $5–$8 per million
Search Queries: $5 per 1,000 queries.
High data volumes (e.g., 1M tokens processed) → may exceed $100 quickly for intensive research projects.
Limited Non-English Support:
While Perplexity’s search can handle multiple languages, the Labs interface and documentation are primarily in English. Global teams may encounter language barriers.
7.4 Perplexity Labs Pricing Summary
Plan
Price
Highlights
Free
$0
Basic search, limited Deep Research, offline model only (r1-1776)
Pro
$20/month ($200/year)
Labs access, GPT-4, Claude 3, Mistral, Llama, up to 10-min autonomous workflows
Enterprise Pro (<250 td=""> </250>
$40/month per seat ($400/year per seat)
Everything in Pro, enterprise security, data retention, seat-based billing
To help you quickly gauge which AI agent best fits your requirements, here’s a consolidated comparison table highlighting each platform’s core functionalities, pros, cons, and pricing tiers.
Aspect
Manus
Flowith
Suna
Convergence AI (Proxy)
Perplexity Labs
Developer
Monica (Butterfly Effect AI)
Flowith (Hangzhou, China)
Kortix AI Corp (USA; Open Source)
Convergence AI (London, UK)
Perplexity AI, Inc. (USA)
Launch Status
March 6, 2025 (Private Beta, invite-only)
Closed Beta (invite-only) 2025
Public Cloud (suna.so) + GitHub (Fully open-source)
• Free: \$0 → Basic search, limited Deep Research • Pro: \$20 (\$200/year) → Labs mode, GPT-4, Claude 3, Mistral, Llama, 10-min workflows • Enterprise Pro (<250 data="" enhanced="" retention="" seat="" security="" td="" year=""> </250>
Table 5: Side-by-Side Comparison of AI Agents
9. Differentiators & Practical Tips
9.1 When to Choose Manus
Your Top Priority: Fully autonomous, end-to-end workflows that involve code execution, web scraping behind paywalls (with user-provided credentials), and generating polished deliverables with minimal supervision.
Team Size: At least 5 seats (Starter Plan \$39/seat/month).
Budget: Willing to invest \$195–\$995/month for 5–25 seats to empower multiple users simultaneously.
Technical Appetite: Users with some comfort configuring API tokens, managing cloud resources, and interpreting intermediate logs.
Avoid if: You need immediate, unrestricted access (private beta invite code required), or your tasks are simple (e.g., light web scraping or small data analyses).
9.2 When to Choose Flowith
Your Top Priority: Projects requiring massive context windows (e.g., entire legal documents, lengthy research manuscripts, large codebases).
Budget: \$0 to \$249.95/month (depending on usage)—exceptionally cost-effective for “context-hungry” workflows.
Use Cases: Long-term projects like writing books, creating 3D game assets, or training domain-specific agents using Knowledge Gardens.
Avoid if: You need straightforward browser automation or prefer a chat-like UI (Flowith’s canvas approach demands time to master).
9.3 When to Choose Suna
Your Top Priority: Transparency & customization—with complete access to source code and the ability to self-host.
Budget: From \$0 (Free) to \$50/month (Custom) for moderate usage (2–6 hours/month).
Use Cases: Research automation, web scraping, data analysis, report generation—particularly in organizations demanding data privacy (e.g., healthcare, finance).
Avoid if: You need advanced multimodal outputs (video, image, 3D) or extended context beyond ~few thousand tokens.
9.4 When to Choose Convergence AI (Proxy)
Your Top Priority: Browser-based task automation with minimal setup—automate form filling, email parsing, or price monitoring via natural language in a zero-code environment.
Budget: \$0 (Free) for testing or \$20/month (Pro) for unlimited tasks.
Use Cases: E-commerce price tracking, auto-login and data extraction, routine report aggregation, email campaign management—especially for teams requiring an easy visual interface.
Avoid if: Your tasks require code execution, data visualization beyond simple CSVs, or extended offline usage.
9.5 When to Choose Perplexity Labs
Your Top Priority: Research-heavy workflows that blend real-time search, AI reasoning, code execution, and data visualization in a single interface.
Budget: \$0 for basic research; \$20/month unlocks labs; \$40/month per seat (Enterprise Pro) for corporate needs.
Use Cases: Market research, competitive intelligence, academic literature reviews, financial analysis, building interactive dashboards for stakeholders.
Avoid if: You prefer a chat-like AI assistant for casual queries (Perplexity’s real power is unlocked only in Labs mode).
10. FAQ: Frequently Asked Questions
Q1. What exactly is the difference between Suna and Suno?
Answer:
Suna (suna.so) is an open-source AI agent by Kortix AI Corp, focusing on web scraping, data analysis, and research tasks.
Suno is a separate platform (e.g., Suno.ai) specializing in AI-generated audio/music.
Bottom Line: Suna automates text-based workflows; Suno generates music/audio content. They serve entirely different domains.
Q2. Can I run Suna locally on my own servers?
Answer: Yes. You can clone the GitHub repository (github.com/kortix-ai/suna) and set up locally by installing prerequisites like Supabase, Redis, and an LLM API key (OpenAI or Anthropic). This self-hosted option ensures data privacy and aligns with organizations that have strict compliance requirements. For detailed instructions, refer to Suna’s README on GitHub.
Q3. How do the agents handle CAPTCHAs or paywalls?
Answer:
Manus: Allows user to supply credentials on paywalled sites; can handle some simple CAPTCHAs if users manually intervene or provide tokens—but may fail for advanced CAPTCHAs.
Flowith: Lacks native CAPTCHA solving; if a paywall interrupts, the workflow stops until manual resolution.
Suna: Basic scraping modules—CAPTCHA pages will cause errors unless bypassed by headless browser extensions; paywalls remain a challenge.
Convergence AI (Proxy): Proxy can handle some login-based paywalls if user provides credentials, but heavy CAPTCHAs remain a blocker.
Perplexity Labs: Primarily research-focused—does not handle CAPTCHAs or paywall navigation; designed to gather open-source data.
Q4. How do I estimate my monthly cost if I choose Perplexity Labs as my research agent?
Answer:
Pro Plan (\$20/month, \$200/year) gets you up to 10 minutes of autonomous Labs runs and access to GPT-4 & Claude 3.
If you integrate via API, you pay per token:
Input Tokens: \$2–\$3 per million
Output Tokens: \$5–\$8 per million
Search Queries: \$5 per 1,000 queries.
For heavy usage (e.g., processing 100 million tokens), expect ~\$200–\$800 in token fees—so factor that into your budget.
Q5. Is there a free trial or free tier for each agent?
Answer:
Manus: No free tier; must be invited to beta.
Flowith: Starter plan is Free with 1,000 one-time credits; supports basic AI models & Agent Neo preview.
Suna: Free tier gives 60 minutes/month of public projects.
Convergence AI (Proxy): Free tier includes 5 sessions/day and 1 saved automation.
Perplexity Labs: Free tier supports basic search and limited Deep Research—Labs mode is disabled.
Q6. Which agent is best for enterprise-grade security and compliance?
Answer:
Manus: Partners with Alibaba’s Qwen AI; rumored to have enterprise features (e.g., dedicated cloud resources, encryption at rest). However, still in closed beta—review NDA terms.
Flowith: Limited public security info; invite-only access suggests some vetting, but no published SOC 2/ISO audits.
Suna: Self-hosted option offers maximum data control; security depends on how well you configure your own infrastructure (e.g., firewalls, encryption).
Convergence AI (Proxy): Uses cloud-based sandbox; does not explicitly advertise security certifications—question if it meets strict GDPR/HIPAA needs.
Perplexity Labs (Enterprise Pro): Offers enhanced security (data retention policies, compliance frameworks) for \$40/month/seat; best choice for teams with strict data governance.
Q7. Can I integrate Claude or GPT-4 into these agents?
Answer:
Manus: Proprietary agent; does not publicly confirm if it uses OpenAI models inside. Likely leverages in-house or partner models (e.g., Alibaba Qwen).
Flowith: Supports multiple models including GPT-4.1-mini, Claude 3.5 Haiku, Gemini 2.5 Flash, Grok 3-mini—you can choose per task.
Suna: You need to supply your API key (OpenAI, Anthropic) during setup—Suna will call those models under the hood for text generation.
Convergence AI (Proxy): Uses a proprietary reasoning engine for automations; does not expose GPT-4 integration to end users.
Perplexity Labs: Pro users can select GPT-4, Claude 3, Mistral, Llama models for both search and labs tasks.
