AI Landscape & Key Players

The Big Picture

The AI ecosystem has distinct layers. Understanding who does what helps you choose the right tools.

┌─────────────────────────────────────────────────────────────┐
│                     APPLICATIONS                            │
│  Claude Code, Cursor, ChatGPT, Copilot, Perplexity          │
├─────────────────────────────────────────────────────────────┤
│                     FRAMEWORKS                              │
│  LangChain, LangGraph, LlamaIndex, CrewAI, Agent SDK        │
├─────────────────────────────────────────────────────────────┤
│                     INFRASTRUCTURE                          │
│  Hugging Face, Vector DBs, W&B, Replicate                   │
├─────────────────────────────────────────────────────────────┤
│                     CLOUD PLATFORMS                         │
│  AWS Bedrock, Google Vertex AI, Azure OpenAI                │
├─────────────────────────────────────────────────────────────┤
│                     FOUNDATION MODELS                       │
│  Claude, GPT, Gemini, Llama, Mistral, DeepSeek, Grok        │
└─────────────────────────────────────────────────────────────┘

Foundation Model Providers

These companies build the core LLMs that power everything else.

CompanyModel FamilyKey StrengthsOpen Source?
AnthropicClaude (Opus 4, Sonnet 4, Haiku 3.5)Safety, coding, long context (200K), agentic useNo
OpenAIGPT (GPT-4o, o1, o3, GPT-4.1)Ecosystem, multimodal, reasoning modelsNo
GoogleGemini (2.5 Pro, 2.5 Flash)Massive context (1M+), multimodal, search integrationNo
MetaLlama (3.1, 4)Best open-source models, run locallyYes
MistralMistral (Large, Medium, Small)European, strong efficiency, open weightsPartially
xAIGrok (3)Real-time X/Twitter data accessNo
DeepSeekDeepSeek (V3, R1)Open-source reasoning model, strong on math/codeYes

Reasoning Models (New Category)

A major trend: models that "think" before answering, spending compute on step-by-step reasoning.

ModelProviderKey Trait
Claude Opus 4AnthropicExtended thinking with visible reasoning traces
o1 / o3OpenAIChain-of-thought reasoning, excels at math and logic
DeepSeek R1DeepSeekOpen-source reasoning model
Gemini 2.5 ProGoogleBuilt-in "thinking" mode

Cloud AI Platforms

Run multiple models through a single cloud provider:

PlatformProviderWhat It Offers
AWS BedrockAmazonAccess Claude, Llama, Mistral, and more via AWS
Google Vertex AIGoogleGemini models + third-party models + MLOps tools
Azure OpenAIMicrosoftGPT models with enterprise Azure security/compliance

When to use these: Enterprise teams that need compliance, VPC deployment, or multi-model access through one billing relationship.


AI Coding Tools

ToolHow It WorksBest For
Claude CodeCLI agent: reads files, writes code, runs commands, iteratesAgentic coding, complex refactors, full autonomy
GitHub CopilotIDE autocomplete + chat, powered by GPT/ClaudeInline suggestions while typing
CursorVS Code fork with deep AI integrationAI-first IDE experience
WindsurfIDE with "Cascade" multi-step AI flowsAgentic IDE with context awareness
ClineOpen-source VS Code extension for agentic codingCustomizable, open ecosystem

How They Compare

FeatureClaude CodeCopilotCursorWindsurf
InterfaceTerminal/CLIIDE pluginFull IDEFull IDE
AutonomyHigh (agent loop)Low (suggestions)MediumMedium-High
Tool useFull (file ops, git, shell)LimitedModerateModerate
ModelClaudeGPT/Claude (configurable)MultipleMultiple
Open sourceSDK is openNoNoNo

AI Infrastructure

ToolCategoryWhat It Does
Hugging FaceModel hubHost, share, and discover models and datasets
Weights & BiasesExperiment trackingLog metrics, compare runs, track model performance
ReplicateModel hostingRun open-source models via API without infra
ModalServerless GPURun GPU workloads on demand (fine-tuning, inference)
Together AIInferenceFast, cheap inference for open-source models
GroqHardware inferenceUltra-fast inference on custom LPU chips

Vector Databases

Essential for RAG systems (see 17 - RAG (Retrieval-Augmented Generation)):

DatabaseTypeDifferentiator
PineconeManaged cloudSimplest to start, scales to billions
WeaviateCloud / self-hostedBuilt-in hybrid search
QdrantCloud / self-hostedRich filtering, high performance
ChromaLocal / embeddedGreat for prototyping, simple API
pgvectorPostgreSQL extensionUse your existing Postgres, no new infra
MilvusSelf-hostedEnterprise scale, open source

Agent Frameworks

FrameworkDescriptionBacked By
Claude Agent SDKOfficial SDK for building agents with ClaudeAnthropic
LangChainMost popular framework, huge integration ecosystemLangChain Inc.
LangGraphGraph-based agent workflows with state managementLangChain Inc.
CrewAIMulti-agent role-based collaborationCrewAI
AutoGenMulti-agent conversation frameworkMicrosoft

Model Comparison Table

As of early 2026 (pricing and capabilities change frequently):

Claude Opus 4Claude Sonnet 4GPT-4oGemini 2.5 ProLlama 4
Context200K200K128K1M+128K
Input $/1M$15$3$2.50$1.25Free (self-host)
Output $/1M$75$15$10$10Free (self-host)
CodingExcellentExcellentVery GoodVery GoodGood
ReasoningExcellent (extended thinking)Very GoodGood (use o3 for reasoning)Very Good (thinking mode)Good
MultimodalVisionVisionVision + AudioVision + Audio + VideoVision
Open SourceNoNoNoNoYes
Best ForDeep reasoning, complex tasksDaily coding, agentsGeneral purpose, ecosystemLong context, multimodalLocal deployment, privacy

How to Choose

Need best coding agent?          → Claude Sonnet 4 or Claude Code
Need deep reasoning?             → Claude Opus 4 or OpenAI o3
Need 1M+ context?                → Gemini 2.5 Pro
Need to run locally / privacy?   → Llama 4 or DeepSeek V3
Need cheapest API?               → Claude Haiku 3.5 or GPT-4o-mini
Need real-time search?           → Perplexity or Grok

Key Trends (2025-2026)

TrendWhat's Happening
Reasoning modelso1/o3, DeepSeek R1, Claude extended thinking — models that "think harder"
Agentic AIModels that use tools, plan, and execute multi-step tasks autonomously
Longer contexts1M+ tokens (Gemini), making RAG less necessary for some use cases
Cheaper inferencePrices dropping 5-10x per year, making AI accessible for more use cases
Open source catching upLlama 4, DeepSeek V3, Qwen 2.5 approaching proprietary model quality
MultimodalVision, audio, video as standard inputs, not special features
On-device AIApple Intelligence, Gemini Nano, Llama on phones and laptops
MCP adoptionStandardized tool interface spreading across the ecosystem
Coding agentsClaude Code, Copilot Workspace, Devin — agents that write and ship code

Resources


Previous: 18 - Evaluations & Testing AI | Next: 20 - AI News & Updates