Claude Code is Anthropic's terminal-native coding agent, also available inside VS Code, JetBrains, a desktop app, and the web. It reads and edits files, runs commands, and works through multi-step engineering tasks directly against a real codebase, with a 1-million-token context window for reasoning across large projects. As of independent Terminal-Bench 2.1 testing published in August 2026, the Claude Opus 5 pairing scores 89.1% at max effort on the harness's 89 real-world terminal tasks, the highest score cited in this comparison; the prior Opus 4.8 pairing scored 78.9% on the same benchmark, showing real, measured improvement between model generations rather than a static number.

Access comes bundled into Claude subscription plans or billed per token through the API, and pricing scales with model choice — Opus 5 runs $5 per million input tokens and $25 per million output tokens at the high end. The tradeoff for that reasoning depth is model choice: unlike Aider, Goose, or OpenHands, Claude Code works only with Anthropic's own models, so switching to a different provider mid-project isn't an option the way it is with the model-agnostic tools on this list.

Best for: Developers who want the deepest reasoning available in a terminal-first workflow and are comfortable committing to one model provider. Pricing: Included in Claude plans, or pay-per-token via API.

Cursor is a full AI-native code editor, built as a fork of VS Code by Anysphere, with AI woven into the core rather than added as an extension. Its Composer feature can plan and execute multi-file changes across a project simultaneously, Background Agents run tasks in parallel, and it supports Claude, GPT, Gemini, and an in-house model called Composer, with full MCP support. The scale here is real: Cursor has crossed $2 billion in annualized recurring revenue in under two years, reported more than 360,000 paying customers by early 2026, and counts Fortune 500 engineering teams among its adopters.

The honest wrinkle is a real, well-documented pricing controversy: in 2025, Cursor shifted from a fixed 500 fast-response allowance to usage-based credits, a change that effectively cut the number of requests a $20 subscription covered to roughly 225 for many users, catching people off guard with unexpected overage charges. CEO Michael Truell issued a public apology over the change, and the credit system remains something power users need to actively monitor rather than a flat, predictable cost. Cursor also requires switching your entire editor rather than adding a plugin, and Neovim and Emacs aren't supported at all.

Best for: Teams that want the deepest agentic IDE feature set and are willing to fully migrate to Cursor's editor. Pricing: Free Hobby tier; Pro $20/mo; higher usage tiers to $200/mo.

GitHub Copilot remains the lowest-commitment option on this list: it's an extension that adds AI completion and agent mode inside whatever editor you already use, including JetBrains IDEs, rather than asking you to switch to a new one. Multiple independent 2026 comparisons describe it as covering "80% of AI coding needs at half the price" of Cursor, and it remains the most widely adopted single tool in this category by raw user count, with tens of millions of developers using it in some form.

Its agent mode is generally considered less deep than Cursor's Composer or Claude Code's autonomous multi-step reasoning, and reviewers consistently frame it as the right choice specifically for budget-conscious developers or teams that don't want the migration cost of switching editors, rather than the top pick for the most complex, multi-file engineering work.

Best for: Developers who want solid AI assistance without leaving their current editor or paying IDE-level prices. Pricing: Free tier; Pro $10/mo.

Aider is a free, open-source (Apache 2.0) terminal coding agent created by Paul Gauthier, with no GUI, no web interface, and no IDE plugin required — it runs entirely against a local Git repository from the command line, editing files directly and auto-committing changes with descriptive messages as it works. It builds a map of an entire repository using tree-sitter syntax parsing, showing function signatures and class definitions without loading every file into context, and an Architect mode pairs a stronger planning model with a cheaper execution model to cut session costs. As of mid-2026, the project reports roughly 44,000-47,000 GitHub stars and 6.8 million PyPI installs, processing an estimated 15 billion tokens a week across active users — real, sustained adoption rather than a launch spike.

It works with more than 100 models, including local ones through Ollama, so switching providers is a config change rather than a migration. The honest caveat, raised by more than one 2026 review, is that Aider is "no longer the benchmark frontrunner it once was" against newer IDE-native agents — it remains a serious, actively maintained tool, but the terminal-only, no-GUI design is a real barrier for developers who want a visual diff review rather than auto-committed changes to inspect after the fact.

Best for: Developers who live in Git and the terminal and want full model flexibility without a subscription. Pricing: Free (open source); pay only your chosen model provider's API costs.

OpenHands (formerly OpenDevin) is a free, open-source, model-agnostic agent built for fuller autonomy than a terminal-only tool: it runs inside a sandboxed Docker environment with its own terminal, code editor, file system, and a real browser, so it can research documentation online while working a task rather than being limited to what's already in the repo. Given a GitHub issue, it plans a solution, writes code, runs the test suite, and opens a pull request end to end, iterating until the task completes or it gets stuck. Reported GitHub star counts vary by source but consistently sit in the tens of thousands, and the project raised an $18.8 million Series A, a real funding signal in a category dominated by either bootstrapped open-source projects or venture-scale IDE companies.

Three deployment paths exist: fully self-hosted and free under MIT license, a bring-your-own-key cloud tier capped at 10 daily conversations on the free plan, or custom Enterprise pricing with VPC deployment and SSO. The sandboxed, browser-equipped design is the clearest differentiator from Aider's simpler terminal-only approach, at the cost of a heavier setup.

Best for: Teams that want a self-hostable agent with real autonomy — including web research and PR creation — rather than a lighter terminal-only tool. Pricing: Free (MIT, self-hosted); free cloud tier (BYOK); custom Enterprise.

Goose is Block's (Square, Cash App, Tidal) open-source coding agent, built in Rust under Apache 2.0 and now governed by the Linux Foundation's Agentic AI Foundation after moving out from under Block's direct internal ownership in 2026. It's genuinely free with no paid tier at all — bring your own LLM key (Claude, GPT, Gemini, or a local Ollama model for zero incremental cost) and pay only what that provider charges. Block runs Goose internally across its own 13,000-plus-person engineering organization, which is a meaningful vote of confidence from the company funding its development, and extensibility runs through MCP, with dozens of community-built extensions covering databases, browsers, and dev tools beyond Block's own official set.

It assumes real comfort with terminals and configuration files rather than offering a point-and-click experience, and MCP-based extensibility means some capabilities (like browsing) are assembled from community extensions rather than shipping natively the way OpenHands' browser does out of the box.

Best for: Developers who want an Apache 2.0, MCP-native agent with full commercial-use freedom and zero vendor lock-in on the model. Pricing: Free (Apache 2.0); pay only for LLM API usage.