Skip to content

Companies and Products Landscape

Overview

The AI Agent field is in a period of rapid development, with major global AI companies and startups all positioning themselves in agent products. This section maps out the major players, products, and technical approaches in both global and Chinese markets, providing a panoramic industry map.

Ecosystem Landscape

graph TD
    A[AI Agent Ecosystem] --> B[Foundation Model Providers]
    A --> C[Agent Platforms/Frameworks]
    A --> D[Vertical Applications]
    A --> E[Infrastructure]

    B --> B1[OpenAI]
    B --> B2[Anthropic]
    B --> B3[Google]
    B --> B4[Chinese Vendors]

    C --> C1[LangChain]
    C --> C2[Microsoft Copilot]
    C --> C3[Coze/Dify]

    D --> D1[Code Agents]
    D --> D2[Customer Service Agents]
    D --> D3[Research Agents]

    E --> E1[Vector Databases]
    E --> E2[Monitoring Tools]
    E --> E3[Sandbox Services]

    style A fill:#e3f2fd

Major Global Companies

OpenAI

Dimension Details
Core models GPT-4o, GPT-4o mini, o3, o4-mini
Agent products Operator (Web Agent), Codex (Code Agent)
Dev tools Assistants API, Function Calling
Protocol Responses API (built-in tools)

Key Products:

  • Operator: An agent that autonomously browses the web to complete tasks
  • Codex: Cloud-based asynchronous code agent
  • ChatGPT + Plugins/GPTs: User-customizable agents
  • Assistants API: API for developers to build agents

Anthropic

Dimension Details
Core models Claude Opus 4, Claude Sonnet 4, Claude Haiku 3.5
Agent products Claude Code (CLI Agent), Computer Use
Dev tools Tool Use API, MCP
Protocol Model Context Protocol (MCP)

Key Contributions:

  • MCP (Model Context Protocol): An open tool integration protocol standard
  • Claude Code: A command-line code agent
  • Computer Use: Enabling Claude to operate computer desktops
  • Extended Thinking: Enhanced agent reasoning capabilities

Google / DeepMind

Dimension Details
Core models Gemini 2.5 Pro, Gemini 2.5 Flash
Agent products Deep Research, Gemini with Extensions
Dev tools Vertex AI Agent Builder
Protocol Agent-to-Agent (A2A) Protocol

Key Contributions:

  • A2A Protocol: An open standard for inter-agent communication
  • Deep Research: Autonomous research agent
  • Vertex AI Agent Builder: Enterprise-grade agent building platform
  • 1M+ context window: Supporting ultra-long context agents

Other Global Companies

Company Core Product Features
Cohere Command R+, Agent API Enterprise RAG and agents
Perplexity Pro Search, Spaces Search-driven research agent
Cognition Devin First AI software engineer
Adept ACT-1 GUI operation agent
MultiOn AgentQ Web automation agent

Chinese Market

Major Companies

Company Core Model Agent Product Features
Zhipu AI GLM-4 AutoGLM, Agent Platform Tsinghua-affiliated, technology leader
Baidu ERNIE ERNIE Agent Platform Search ecosystem integration
ByteDance Doubao Model Coze Low-code agent platform
Alibaba Qwen Tongyi Agent, Bailian Platform Cloud service integration
Moonshot AI Moonshot Kimi Long context, search-enhanced
DeepSeek DeepSeek V3/R1 Open-source models High cost-performance, open-source

Coze

ByteDance's low-code agent platform:

  • Visual orchestration: Drag-and-drop agent workflow design
  • Plugin ecosystem: Rich third-party plugins
  • Multi-channel publishing: Supports Feishu, WeChat, web, etc.
  • Chinese market optimization: Localized experience

Bailian Platform (Alibaba)

Alibaba Cloud's agent development platform:

  • Deep integration with Alibaba Cloud ecosystem
  • Supports the full Qwen model series
  • Enterprise-grade security and compliance
  • Unified RAG, Agent, and workflow platform

Product Comparison

Agent Capability Comparison

Capability OpenAI Anthropic Google Zhipu ByteDance
Code generation Codex Claude Code Gemini Code CodeGeeX -
Web operation Operator Computer Use Extensions AutoGLM Coze
Research analysis Deep Research - Deep Research - -
Tool protocol Function Call MCP A2A - Coze Plugin
Multimodal GPT-4V Claude Vision Gemini GLM-4V -

Positioning Comparison

graph LR
    subgraph Developer-oriented
        A[OpenAI API]
        B[Anthropic API]
        C[Google Vertex]
    end

    subgraph Platform-oriented
        D[Coze]
        E[Dify]
        F[Bailian]
    end

    subgraph End User-oriented
        G[ChatGPT]
        H[Claude]
        I[Kimi]
    end

Protocol Standards Competition

MCP vs. A2A

Dimension MCP (Anthropic) A2A (Google)
Positioning Agent-to-tool connection Agent-to-agent communication
Relationship Complementary Complementary
Openness Open-source Open-source
Ecosystem Widely adopted Growing
Technology JSON-RPC + SSE HTTP + JSON

The two protocols actually solve different problems and can coexist:

  • MCP: How agents call external tools and data sources
  • A2A: How different agents communicate and collaborate

Investment and Market

Major Funding Rounds (2024-2025)

Company Funding Amount Valuation Date
OpenAI $6.6B $157B 2024.10
Anthropic $4B (Amazon) $60B+ 2024-2025
Cognition (Devin) $175M $2B 2024.03
Zhipu AI ¥3B ¥20B 2024

Trend Assessment

  1. Model layer convergence: Capability gaps between providers are narrowing
  2. Differentiation through agents: Agent capabilities becoming a new competitive dimension
  3. Protocol standardization: MCP/A2A and similar protocols driving ecosystem interoperability
  4. Vertical deepening: General agents moving toward vertical domain specialization
  5. Open-source catching up: Open-source community rapidly following closed-source capabilities

References

  1. OpenAI. "Introducing Operator." 2025.
  2. Anthropic. "Model Context Protocol." 2024.
  3. Google. "Agent-to-Agent Protocol." 2025.
  4. Various company announcements and product pages.

Cross-references: - Open-source ecosystem → Open-Source Ecosystem - Agent frameworks → LangChain and LangGraph - Market analysis → Market Analysis and Use Cases

评论 #