The AI coding assistant market has three main layers: the model providers (OpenAI, Anthropic, Google, xAI), the AI coding assistant products (GitHub Copilot, Cursor, Windsurf), and the cloud coding agents (Copilot coding agent, Claude coding agent, Codex agent).

It can seem like there are dozens of competing AI tools — but most of them fit neatly into the AICA Onion layers. Once you know the layers, you can place any new tool quickly: is it a model? An assistant? An agent? An orchestrator?

About this comparison

This site specialises in GitHub Copilot — you’ll find deep coverage of its features, agents, enterprise controls, and practical patterns throughout. For other tools, we offer an honest structural map: where each one sits in the AICA Onion layers, what makes it distinct, and where to go for the full story. We do not maintain a feature-by-feature matrix for competitors — that’s each vendor’s job, and any matrix we built would be outdated within weeks.

Use this page to orient yourself in the landscape. Then visit each tool’s official documentation for current pricing, features, and capabilities.

This landscape moves fast — models update every 2-4 weeks, and products ship new features monthly. The comparison below reflects the state as of the last_reviewed date in the page header. Always check official docs for current information before making a purchasing or adoption decision.

Model providers

The models powering every AI coding assistant come from a small number of providers. You don’t interact with these directly through most assistants, but knowing who makes the underlying engine helps you understand why different assistants behave differently.

ProviderKey modelsNotes
OpenAIGPT-5.4, GPT-5.4 miniThe original mainstream LLMs; power Copilot’s default
AnthropicClaude Sonnet 4.5, Claude Opus 4.6Known for large context windows and coding quality
GoogleGemini 2.5 ProStrong long-context and multi-modal capabilities
xAIGrok Code Fast 1Code-focused, fast inference
MetaLlama familyOpen weights — used in self-hosted setups

This table reflects the landscape at time of writing. The model market moves fast — check each provider’s documentation for the latest.

AI coding assistants

These are the Layer 1 products you install and interact with every day. Each one wraps one or more models and adds a UI, context management, guardrails, and billing.

ProductMade byKey differentiatorLearn more
GitHub CopilotGitHub / MicrosoftMulti-model, GitHub-native, agent mode + coding agent + orchestrationCopilot docs
CursorAnysphereDeep codebase indexing, AI-native IDE (VS Code fork), Composer agentCursor docs
WindsurfCodeiumMCP-first architecture, Cascade agent, browser previewWindsurf docs
Amazon Q DeveloperAWSAWS ecosystem expert, code transformation agents, security scanningAmazon Q docs
JetBrains AI AssistantJetBrainsNative integration across all JetBrains IDEs; local model supportJetBrains AI docs
TabnineTabnineCompletion-focused, privacy-first, 40+ IDEs, VPC/on-premise deploymentTabnine docs

Cloud coding agents

Cloud coding agents are a newer and fast-moving category. Unlike agent mode (which runs locally in your IDE), these agents run in the cloud, work asynchronously, and hand you back a pull request when they’re done.

AgentMade byHow you access it
Copilot Coding AgentGitHubAssign a GitHub Issue to Copilot
Claude Coding AgentAnthropicVia GitHub Copilot (third-party, public preview) or direct API
Codex AgentOpenAIVia GitHub Copilot (third-party, public preview) or direct API
DevinCognitionStandalone product
SWE-agentPrinceton NLPOpen source research agent

Third-party agents in GitHub Copilot are in public preview.

How the tools compare

This table maps each tool across six key dimensions. It is a high-level snapshot, not an exhaustive feature matrix — check each tool’s official docs for current detail.

DimensionCopilotCursorWindsurfAmazon QJetBrains AITabnine
IDE supportPlugin (all major IDEs)Standalone IDEStandalone IDECloud + consolePlugin (all JetBrains IDEs)Plugin (40+ IDEs)
Cloud agent✅ Copilot Coding Agent❌ No❌ No✅ Code transform agents❌ No❌ No
Local agentic mode✅ Agent mode✅ Composer✅ Cascade❌ No✅ Agent mode❌ No
Model choice✅ 10+ models (4 providers)✅ Multiple providers✅ Multiple providersAWS-focused✅ Cloud + local (Ollama)❌ Tabnine models only
Enterprise controls✅ Content exclusion, data residency, audit logs, IP indemnity⚠️ Limited⚠️ Limited✅ AWS-specific✅ Local model option✅ VPC/on-premise deployment
Local model support❌ No❌ No❌ No❌ No✅ Yes (Ollama, LM Studio)⚠️ Partial (VPC)

This is a high-level snapshot verified against official documentation as of 2026-04-04. Some cells marked ⚠️ reflect uncertainty or limited public documentation — check vendor sources directly. The landscape changes monthly; do not rely on this table for procurement decisions without verifying with the vendor.

GitHub Copilot’s unique position

GitHub Copilot is unusual because it operates at multiple layers simultaneously. It’s a Layer 1 AI coding assistant, it has its own Layer 2 agents (agent mode and the Copilot coding agent), and it acts as a Layer 3 orchestrator routing to third-party agents. No other product in this table spans as many layers.

For a detailed look at what this means in practice for enterprise teams — including content exclusions, data residency, policy controls, audit logs, and IP indemnity — see Enterprise & Trust.

How to read new tools

💡 When you encounter a new AI coding tool, ask:

  1. Is it a model (predicts tokens) or a product (wraps models)?
  2. Does it act autonomously (agent) or respond to prompts (assistant)?
  3. Can it route to other agents (orchestrator)?

The answers place it in the AICA Onion immediately.

Official documentation

Each tool’s official docs are the authoritative source for current features, pricing, and configuration:

Dive deeper

📍 The Full Landscape · Spans All Layers · Back to the AICA Onion

What’s next?

Updated: