What GitHub Copilot Does
GitHub Copilot is the AI coding assistant that defined the category. Launched in 2021 as a collaboration between GitHub and OpenAI, it started as a single-purpose autocomplete tool that suggested code as you typed. Four years later, it has transformed into something far more ambitious: a multi-model, agentic coding platform that reaches across your entire development workflow — from inline completions in your editor to autonomous agents that turn GitHub issues into reviewed pull requests while you sleep.
Inline Completions and Multi-Model Support
The inline completions remain Copilot's bread and butter, and they are still the fastest and most accurate in the industry. Suggestions appear almost instantly, often before you finish typing a function signature. The acceptance rate is high because the model has been trained on an enormous corpus of code and benefits from deep integration with the GitHub ecosystem. For many developers, these completions alone justify the subscription — the small time savings on each suggestion compound into hours saved per week.
Multi-model support arrived as one of the most significant additions. Developers can now switch between models from OpenAI, Anthropic, and Google directly within Copilot Chat, matching the model to the task at hand. Claude for nuanced reasoning about complex architecture, GPT for speed on routine questions, Gemini for its large context window — having model choice within your existing workflow means you do not need to switch tools to access different AI capabilities. This is a meaningful advantage over AI-first editors that lock you into a single provider.
Agent Mode and Coding Agent
Agent mode represents Copilot's push into agentic development. Unlike traditional chat or inline edits, agent mode can independently translate ideas into code, identifying necessary subtasks and executing them across multiple files. It runs terminal commands, fixes lint errors, and iterates autonomously until the task is complete. For straightforward tasks — adding error handling to all API endpoints, creating tests for a module, refactoring a function — agent mode works well enough that you do not need a separate tool. However, complex multi-file refactoring involving ten or more files with architectural implications is where Cursor and Claude Code still pull ahead. Copilot's agent tends to make more mistakes on tasks at that scale.
The coding agent is perhaps the most forward-looking feature. You can assign a GitHub issue directly to Copilot, and it will work in the background — spinning up a secure development environment powered by GitHub Actions, making the required changes, running tests, performing its own code review, and opening a pull request for you to review. For well-scoped, single-component tasks, this works reliably. It excels at bug fixes, test additions, documentation improvements, and small feature implementations. The agent even runs code scanning, secret scanning, and dependency vulnerability checks before opening the PR, catching security issues before they reach human review.
Copilot Fusion and GitHub Universe 2026
At GitHub Universe in October 2026, GitHub made Copilot Fusion generally available. Instead of choosing a model, you pick Copilot Fusion and an effort level (Standard, High or Max), and Copilot decides which models to use for each prompt: one model on its own, a cheaper model whose draft is escalated to a stronger one if it fails a quality check, or one model drafting while another reviews. Fusion is billed for each model it uses at that model's standard rate, with a limited-time 30% discount on paid plans, and works on all Copilot plans in VS Code, JetBrains IDEs, Copilot CLI and the GitHub Copilot app. Administrators decide which models Fusion may use and can make it the default. GitHub cites an Artificial Analysis Coding Agent Index score of 60 for Fusion Max; that is GitHub's own claim.
An opt-in experimental preview also lets Fusion route between local on-device models and cloud models in Copilot CLI, the Copilot app and VS Code. On Windows, Copilot can download MAI Code 1.1 Flash through Windows ML; Copilot CLI on Windows, macOS and Linux can use a manually configured MAI Code 1.1 Flash or Qwen3.8 Flash Next through a local OpenAI-compatible server. Routing decisions still happen in the cloud, so sessions are not offline.
Pricing also changed for large companies. Enterprises working with a GitHub or Microsoft account team can now get Copilot included with GitHub Enterprise Cloud for every licensed user, with AI usage metered from the first request against an enterprise budget instead of monthly per-seat credits; code completion and next edit suggestions keep working when the budget runs out. Users in such an enterprise without a GitHub Enterprise license can get Copilot Standalone for $10 per user per month. Self-serve availability is rolling out in stages, and Copilot Business and Copilot Enterprise remain on sale.
IDE Support and Code Review
IDE support is unmatched in the industry. Copilot works natively in VS Code, Visual Studio, JetBrains IDEs, Eclipse, Xcode, and Neovim. Agent mode has expanded beyond VS Code into JetBrains, Eclipse, and Xcode as well. No competing tool comes close to this breadth. For teams with mixed editor preferences — frontend developers in VS Code, backend engineers in IntelliJ, mobile developers in Xcode — Copilot is the only AI assistant that works across the entire team without compromise.
Code review reached sixty million reviews by early 2026, growing tenfold since its launch. The agentic architecture means it now gathers full repository context before commenting — it understands how a change interacts with the broader codebase rather than just looking at the diff. In the majority of reviews it surfaces actionable feedback, and when it has nothing useful to add, it stays silent rather than generating noise. Importantly, Copilot's review counts as a comment rather than an approval, so it never blocks merges or replaces human review in protected branch policies.
Pricing and Limitations
Copilot now bills AI usage in GitHub AI Credits, where one credit is worth one cent, and every plan includes a monthly allowance. Code completions and next edit suggestions do not use credits and are unlimited on paid plans; chat, agent mode, code review, Copilot in the cloud and Copilot CLI draw credits at a rate that depends on the model and the task. The Free plan includes 2,000 completions a month plus limited chat and agent use with entry-level models. Pro costs $10 a month and includes $15 of credits, Pro+ costs $39 with $70 of credits and adds premium models such as Claude Opus, and Max costs $100 with $200 of credits and priority access to new models. For organizations, Copilot Business costs $19 and Copilot Enterprise $39 per granted seat per month, with 1,900 and 3,900 credits per user pooled across the organization. When an allowance runs out, you can wait for the monthly reset, set a budget for additional paid usage, or switch to a cheaper model.
Copilot's weaknesses are real but not disqualifying. Complex multi-file editing is weaker than Cursor's Composer. Deep reasoning about architecture is less insightful than Claude. The coding agent struggles with multi-component changes that require understanding ripple effects across a codebase. Heavy users of agent mode can use up their monthly credits before the month ends. And like all AI coding tools, Copilot hallucinates — it will suggest plausible but incorrect API calls, reference nonexistent libraries, or generate code with subtle logic errors that pass a cursory glance.
The Bottom Line
But Copilot is still the best value proposition in AI-assisted development. At ten dollars per month with unlimited autocomplete, a monthly credit allowance for multi-model chat, agent mode, coding agents and code review, and the broadest IDE support available, it delivers more per dollar than any competitor. The GitHub integration — issues, pull requests, Actions, code review — creates a flywheel that no standalone editor can replicate. Four years in, Copilot has evolved from a novelty into essential developer infrastructure. It may not be the most powerful tool in any single dimension, but it is the one most developers should start with, and many will never need to leave.





