Skip to content

The AI Landscape: Companies, Models & Tools

📖 5 min read getting-startedreference
How the major AI labs are structured and where every other tool fits — Anthropic, OpenAI, Google, DeepSeek, plus open weights, wrappers, and infrastructure.

A map of the modern AI ecosystem: how each major lab is organized, and where the tools you’ve heard of (Cursor, Perplexity, LangChain…) actually fit. If you just want to pick a model, jump to the Models Decision Guide or the comparison on the home page.


How the Major AI Companies Are Structured

Each major lab has a full vertical stack - a research org, a model family, products built on those models, and distinct tiers you choose between. Understanding this helps you know exactly what you’re paying for and why the same company can have five different products.

Anthropic

Founded 2021 · Safety-focused · San Francisco

Model Family: Claude

Products:

  • claude.ai - Chat interface (web & mobile)
  • Anthropic API - Direct model access for developers
  • Claude Code - Agentic CLI for complex coding tasks and refactoring
  • Claude Cowork - AI for knowledge work (Jan 2026 research preview); give Claude a goal and it works autonomously on your computer, files, and apps
  • Claude Dispatch - Mobile-to-desktop workflow layer (Mar 2026); text tasks from your phone, Claude executes them on your desktop with full computer control; Max/Pro subscribers, macOS
  • Claude for Teams / Enterprise - Business plans with data privacy and compliance
  • Powering other apps - Cursor, Windsurf use Claude under the hood

Model Tiers (cheapest → most capable):

Claude Opus 4.8 $5/25 per 1M tokens
1M context reasoning, coding, writing, analysis, vision

Most capable Claude (May 2026). Best for complex reasoning and agentic coding. Adaptive thinking. Fast Mode $10/$50.

Claude Opus 4.8 (Thinking) $5/25 per 1M tokens
1M context reasoning, coding, writing, analysis, vision, design

Top-ranked on Design Arena. Thinking mode enabled.

Claude Sonnet 4.6 $3/15 per 1M tokens
1M context coding, reasoning, writing, analysis, vision

Best balance of speed & quality. Default pick.

Claude Haiku 4.5 $1/5 per 1M tokens
200K context classification, routing, summarization, vision

Ultra-fast, cheapest Claude.

All tiers share 1M token context (Opus 4.8 and Sonnet 4.6); Haiku 4.5 has 200K. See the Claude section for complete ecosystem details.


OpenAI

Founded 2015 · Microsoft-backed · San Francisco

Model Families:

  • GPT-5.5 - General purpose (including instant variant for cost-efficiency)
  • o3 - Test-time compute reasoning (spends tokens on hidden “thinking” before answering)

Products:

  • ChatGPT - Chat interface (web, mobile, desktop); includes web search, code execution, image generation
  • OpenAI API - Direct model access for developers
  • GitHub Copilot - Code-powered IDE integration (powered by OpenAI models)
  • Image-2 - Image generation (inside ChatGPT Plus)
  • Sora - Video generation (ChatGPT Pro)
  • ChatGPT Enterprise - Business plan, data privacy, org-wide deployment

Model Tiers (cheapest → most capable):

GPT-5.5 $5/30 per 1M tokens
1M context general, coding, writing, reasoning, vision

Flagship. Reasoning levels none→xhigh. Strong all-around.

GPT-5.4 $2.5/15 per 1M tokens
1M context general, coding, writing, reasoning, vision

Affordable professional tier. Near-flagship capability.

GPT-5.4 mini $0.75/4.5 per 1M tokens
400K context general, coding, computer-use, subagents

Strong mini for coding & agents. Fast.

GPT-5.4 nano $0.2/1.25 per 1M tokens
400K context general, classification, routing

Fastest, cheapest. Ideal for high-throughput.

o3
128K context reasoning, math, science, coding

Dedicated reasoning model. Spends tokens on hidden thinking. 87% cheaper than o1.

$2/$8 per 1M

o3 allocates compute at inference time (“thinking” before answering) — slower but smarter on math, coding, reasoning.


Google DeepMind

Founded 1998 (Google) · Alphabet subsidiary · Mountain View

Model Family: Gemini + Imagen + Veo (video)

Products:

  • Gemini.google.com - Chat interface (web & mobile); includes Deep Research mode for extended exploration
  • Google AI Studio - Free API access for developers
  • Vertex AI - Enterprise API with Google Cloud integration
  • Gemini in Workspace - AI inside Docs, Gmail, Sheets, Slides
  • Imagen / ImageFX - Image generation with editing
  • Veo 3.1 - Video generation (cinematic quality)
  • Gemini on Android - Native assistant replacing Google Assistant

Model Tiers (cheapest → most capable):

Gemini 3.1 Pro $2/12 per 1M tokens
1M context reasoning, research, vision, long-context, video

Flagship Gemini. Best context window, excellent multimodal. Prompts >200K billed $4/$18.

Gemini 3.5 Flash $1.5/9 per 1M tokens
1M context reasoning, coding, vision, speed

Fast Gemini. $0.15/M cached input (90% off). Free tier on AI Studio.

Gemini 3.1 Pro’s 1M-token context enables entire research papers in a single request. See the DeepMind section for complete ecosystem details.


DeepSeek (High-Flyer)

Founded 2023 · Hangzhou, China · Open weights (MIT license)

Model Family: DeepSeek V4 (Flash + Pro; reasoning via built-in thinking mode)

Products:

  • chat.deepseek.com - Chat interface (free)
  • DeepSeek API - Direct access with enterprise pricing
  • Open weights - Run locally via Ollama, Hugging Face, or LM Studio (fully self-hosted)

Model Tiers (cheapest → most capable):

DeepSeek V4 Flash $0.14/0.28 per 1M tokens
1M context routing, classification, general, reasoning

Cost leader. MIT license. FREE on OpenCode.

DeepSeek V4 Pro $0.435/0.87 per 1M tokens
1M context reasoning, coding, general, design

Premium tier. Thinking mode default. 75% price cut now permanent (announced May 22, 2026).

MIT license = free for commercial use. Most cost-effective by far (15-100x cheaper). See the DeepSeek section for complete ecosystem details.


How Other Tools Fit Into This Picture

Not every AI tool builds its own model. Most sit on top of the big three, or rely on the open-source community. Here’s where everything else fits:

Open Weights - No product, just models

Meta (Llama), Mistral, Google (Gemma), Microsoft (Phi-4), Yi (01.AI)

These companies release model weights publicly but don’t operate a major consumer product. They build the engine and let the community use it.

How you access them:

  • Ollama (run locally)
  • Hugging Face
  • Replicate
  • Together AI
  • LM Studio

Sits on top of the big three

Perplexity, Microsoft Copilot, GitHub Copilot, Cursor, Windsurf, Aider

These products don’t build their own foundation models. They call the OpenAI, Anthropic, or Google APIs and wrap them in specialized experiences (often agentic ones).

  • Perplexity → Uses Claude + GPT-5.5 + proprietary search and synthesis
  • Microsoft Copilot → Powered by OpenAI GPT-5.5 with web integration
  • GitHub Copilot → Powered by OpenAI models (code-optimized)
  • Cursor → Agentic IDE using Claude Sonnet or GPT-5.5 (you choose); 78% SWE-bench
  • Windsurf → Agentic IDE with cascading AI; uses Claude or Codeium models; 75% SWE-bench
  • Aider → Git-native AI coding agent; terminal-based, works with Claude or GPT-5.5

Infrastructure & Routing

LangChain, LlamaIndex, OpenRouter, CrewAI

These tools don’t provide AI themselves - they help you build systems that connect to and orchestrate multiple models and data sources.

  • OpenRouter → single API to route between 200+ models
  • LangChain → framework to chain LLM calls and tools
  • LlamaIndex → connects your data to any LLM
  • CrewAI → orchestrates multiple AI agents working together