Code Assistants

The original mainstream code completion tool, now expanded into a full multi-model agent platform owned by Microsoft. Covers more IDEs than any competitor and offers GitHub-native features like Coding Agent for cloud-side PR work.

• Models: GPT-5.5, GPT-5 mini, GPT-5-Codex, GPT-4.1, Claude Opus 4.7, Sonnet 4.5, Gemini 2.0 Flash
• IDEs: VS Code, Visual Studio, full JetBrains suite, Neovim/Vim, Xcode (native), Eclipse, Azure Data Studio
• Agent Mode (GA on VS Code + JetBrains) for autonomous iteration; Coding Agent runs cloud tasks
• MCP registry for org-controlled allowlisting of trusted MCP servers
• Pricing: Free (~2K completions/mo), Pro $10/mo, Pro+ $39/mo, Business $19/seat, Enterprise $39/seat
• Business/Enterprise: ZDR with OpenAI and Anthropic, IP indemnity, no training on customer data

Limitations: Premium-request quotas are tightening — Pro/Pro+ sign-ups paused April 20, 2026 ahead of a token-based billing shift. Frontier models like GPT-5.5 and Claude Opus 4.7 carry 7.5x premium-request multipliers, burning quota fast. Chat surfaces limited to VS Code, JetBrains, and Visual Studio.

A VS Code fork built around AI-first interaction. Cursor Tab provides sub-100ms inline completions while Composer 3 handles long-running, multi-file agent tasks. Default agent model is Claude Sonnet 4.6, with frontier-model switching per conversation.

• Models: Claude Sonnet 4.6, Claude Opus 4.6/4.7, GPT-5.3/5.4, Gemini 3 Pro, plus in-house Composer and cursor-tab-3
• Composer 3 runs multi-hour, multi-file agent sessions with parallel agent support
• Repository-wide indexing and embedding-based search built in
• Full MCP support including local MCP tools usable in strictest privacy modes
• Privacy Mode, Privacy Mode (Legacy), and Ghost Mode (on-device) for tiered ZDR
• Pricing: Hobby Free, Pro $20/mo, Pro+ $60/mo, Ultra $200/mo, Teams $40/seat

Limitations: Being a VS Code fork, you can't use the official VS Code Marketplace — uses Open VSX, with most popular extensions ported but some proprietary ones missing. Repo indexing is CPU-heavy on initial setup of large monorepos. Not available outside the Cursor app (no JetBrains, no Vim).

Originally Codeium, Windsurf is now owned by Cognition AI (makers of Devin) after a 2025 ownership saga that saw Google poach the founding team and Cognition acquire the remaining IP. Ships its own VS Code-fork editor plus plugins for 40+ IDEs — the broadest non-Copilot IDE coverage available.

• Models: Claude Sonnet 3.5/4.x, GPT-5.4, Gemini 3 Pro, plus in-house SWE-1.5 at ~950 tok/s on Cerebras
• Cascade agent with Code/Chat modes, voice input, checkpoints, real-time linter awareness
• Wave 13 features: parallel multi-agent sessions (up to 5), Arena Mode A/B model comparisons, Plan Mode
• IDEs: Windsurf editor + plugins for full JetBrains suite, Vim/Neovim, Emacs, VS Code, Eclipse
• Native MCP integration with stdio/HTTP/SSE transports; admin allowlists for Teams
• Pricing: Free (unlimited Tab + limited Cascade), Pro $15/mo, Teams (admin), Max $200/mo

Limitations: Ownership transition is still settling — Cognition is gradually integrating Windsurf with Devin. Some users report inconsistency in roadmap signals after the founding-team exit. JetBrains plugin is solid but the flagship Cascade agent experience is best in the Windsurf editor itself.

AWS's coding agent, built on Anthropic Claude models running on Bedrock. Distinguishes itself with deep AWS-specific capabilities: CDK/CloudFormation generation, IAM policy authoring, cost analysis, and the headline /transform agent for multi-version Java and .NET upgrades.

• Models: Claude family on Bedrock (managed by AWS); not user-selectable
• Slash commands: /dev, /transform, /review, /security, /test, /doc
• IDEs: VS Code, JetBrains suite, Visual Studio, Eclipse, AWS Console, Q CLI for terminal/AWS ops
• MCP support added to Q CLI and IDE plugins for custom tools/data sources
• Pro tier ($19/seat) includes data isolation, IP indemnity, IAM Identity Center SSO
• Code transformation: 4,000 LOC/month per Pro seat included for Java upgrades

Limitations: No model choice — AWS picks the model behind the scenes. Tightest fit when you live in the AWS ecosystem; less compelling for Azure/GCP-only shops. Context size not user-configurable. The agent is solid but doesn't yet match Claude Code, Cursor, or Windsurf on long-horizon autonomous work.

Open-source VS Code extensions that add agentic coding using your own API keys (BYOK). Cline (formerly Claude Dev) is the original; Roo Code is a community fork that ships features faster. The extensions are free; you pay only the model provider directly — no markup, no seat licenses.

• Models: Anthropic, OpenAI, OpenRouter, Bedrock, Vertex, Gemini, DeepSeek, Mistral, Ollama, LM Studio, any OpenAI-compatible endpoint
• Plan/Act modes (Cline) and Custom Modes like Architect/Debug/Ask (Roo Code)
• First-class MCP support — install/manage MCP servers from the extension
• Browser use via Puppeteer; checkpoint-based diff approval workflow
• Roo Code adds: codebase indexing, multiple parallel agents, per-mode model selection
• Privacy: runs locally; code only leaves via the API endpoint you choose; air-gappable with Ollama

Limitations: VS Code-only (works in any VS Code-compatible editor). No JetBrains or Visual Studio support. No managed inference — you'll set up and pay for API access yourself. Real-time token cost tracking helps, but unmonitored agent runs on Opus or GPT-5 can rack up real money fast. Setup is more involved than installing a polished commercial tool.

OpenAI's open-source terminal coding agent, written in Rust. Pairs GPT-5.5, GPT-5-Codex, and o-series reasoning models with local file access. Also ships as VS Code and JetBrains IDE extensions that share the same config. Included in ChatGPT Plus/Pro/Business/Enterprise — no separate seat license.

• Models: GPT-5.5 (flagship), GPT-5-Codex (optimized for speed), codex-mini (low-latency Q&A), o3/o4-mini reasoning
• Open-source Rust CLI + VS Code extension + JetBrains plugin (IntelliJ, PyCharm, WebStorm, Rider)
• MCP support in both CLI and IDE — shared config across surfaces
• Cloud sandbox execution for safe autonomous runs
• Included in ChatGPT Plus $20/mo, Pro $100/mo (2x usage promo through May 2026), Pro $200/mo (20x usage)
• 2M+ weekly active users; Codex Security agent for vulnerability scanning

Limitations: OpenAI models only — no Anthropic or Google model switching. Token-based billing shift in progress (moved from credits to per-token in April 2026), so cost visibility is still settling. No inline tab autocomplete — agent/chat interaction only, similar to Claude Code. IDE extensions are newer and less mature than the CLI.

A fully open-source, terminal-first AI coding agent with 140K+ GitHub stars and 6.5M+ monthly developers. Supports 75+ model providers via BYOK — Anthropic, OpenAI, Gemini, Bedrock, Ollama, and any OpenAI-compatible endpoint. Plan/Build dual-agent system, LSP integration, and MCP support.

• Models: Any provider — Anthropic, OpenAI, Google Gemini, Bedrock, Ollama, LM Studio, 75+ supported
• Plan/Build dual agents: Plan drafts a strategy, Build executes — review before any files change
• Terminal TUI + desktop app + IDE extension — keyboard-driven workflow
• Built-in LSP diagnostics surfaced to the agent for real-time error awareness
• MCP support for GitHub, PostgreSQL, filesystem, and custom tools via opencode.json
• Free and open source — no subscription, no account required, pay only your model provider

Limitations: No managed inference — you set up and pay for API access yourself. No inline tab autocomplete (agent interaction only). Newer than Cline/Roo Code with a faster-moving but less stable plugin ecosystem. Desktop app and IDE extension are less mature than the terminal TUI. No enterprise tier with IP indemnity or SSO.

Google's agent-first IDE, launched November 18, 2025 alongside Gemini 3. A VS Code fork built around an "Agent Manager" surface where you spawn and orchestrate multiple agents working asynchronously across different workspaces. Built-in browser preview lets agents drive a real browser and produce verifiable artifacts (screenshots, traces, test results).

• Models: Gemini 3 / 3.1 Pro (default), Claude Sonnet 4.6, Claude Opus 4.6/4.7, GPT-OSS 120B
• Manager Surface — spawn, observe, and orchestrate parallel async agents across workspaces
• Editor surface — traditional VS Code experience with tab completion and inline commands
• Built-in browser preview with artifact generation for agent self-verification
• MCP support added in early 2026 for GitHub, databases, APIs, and custom tools
• IDEs: standalone Antigravity app on macOS, Windows, Linux (no extension form)
• Pricing: Free public preview (all models, rate-limited), AI Pro $20/mo, AI Ultra $249.99/mo

Limitations: Standalone app only — no JetBrains, Visual Studio, Vim, or Xcode support. Public preview status means rapid changes and stability hiccups. Users report rate-limit tightening and opaque credit pricing on paid tiers post-launch. Open VSX marketplace (not the official VS Code marketplace) means some proprietary extensions are unavailable. Less mature as a general-purpose editor than Cursor or Windsurf.