GitHub’s AI pair programmer, built by Microsoft and GitHub on top of the largest code-and-repo graph in the world. Ships as a first-party extension inside VS Code, Visual Studio, JetBrains, Xcode, and Neovim. The GA Coding Agent runs asynchronously on GitHub Actions to turn issues into pull requests.
Claude Opus 4.7 went generally available inside Copilot on April 16, 2026, within hours of Anthropic’s release. GPT-5.5 followed on April 24 for Pro+, Business, and Enterprise users at a 7.5x premium-request multiplier. On May 19, Gemini 3.5 Flash became generally available in GitHub Copilot, making Copilot an even clearer multi-model coding surface.
On May 20-21, GitHub added semantic issue search, smarter auto model routing, Eclipse source transparency, GitHub-owned report URLs, and web model cleanup. The important nuance is that all Gemini models were removed from Copilot Chat on GitHub.com specifically, so Copilot model availability must now be checked per surface rather than assumed globally.
On May 18, GitHub shipped a cluster of Copilot agent-control updates: remote control for Copilot CLI sessions reached GA, the Copilot Spaces API reached GA, lower-cost cloud-agent model options arrived for simple tasks, and repository cloud-agent configuration auditing entered public preview.
On April 24, 2026, GitHub’s Copilot interaction-data policy took effect for Free, Pro, and Pro+ accounts: prompts, outputs, code snippets, and associated context may be used for model training unless individual users opt out. Copilot Business and Enterprise remain excluded from that training path.
On May 14, 2026, GitHub put the Copilot App into technical preview. The desktop app starts agent sessions from GitHub issues, pull requests, prompts, or prior sessions; keeps each task in its own branch/files/conversation space; and lets users validate, review, open PRs, and use Agent Merge from the same workflow.
Also on May 14, The Verge reported that Microsoft is canceling most internal Claude Code licenses in its Experiences and Devices division and pushing engineers toward GitHub Copilot CLI by the end of June. Treat this as a reported internal Microsoft tooling shift, not a public Claude Code deprecation; the buyer signal is that Microsoft is converging its own agentic CLI work around Copilot.
On May 6, 2026, ServiceNow Build Agent reached GitHub Copilot, Cursor, Windsurf, and Claude Code, giving ServiceNow developers a path to work from Copilot while retaining ServiceNow platform context and governance.
On May 5, 2026, GitHub brought secret and dependency scanning into MCP developer workflows. Secret scanning through the GitHub MCP Server is GA, dependency scanning is in public preview, and the Defender for Cloud code-to-cloud integration is GA.
On May 1, 2026, GitHub announced that GPT-5.2 and GPT-5.2-Codex will be retired across most Copilot experiences on June 1, pushing users to GPT-5.5 and GPT-5.3-Codex. Enterprise admins should confirm model policies before the cutoff.
On April 27, 2026, GitHub announced that Copilot will move to usage-based billing on June 1. Premium requests are being replaced by GitHub AI Credits, and private-repo Copilot code review will begin consuming GitHub Actions minutes.
Also on April 27, GitHub said Copilot cloud agent now starts 20% faster when repositories use Actions custom images. On April 29, the rise of agent skill libraries showed the workflow layer Copilot now competes in: reusable instructions, agent templates, and versioned team habits rather than one-off prompts.
The April 26 news scan added five Copilot-specific updates: BYOK in VS Code, cloud-agent metrics, Jira-agent controls, PR chat improvements, and web debugging. Together, they make Copilot look less like autocomplete plus chat and more like an admin-governed coding-agent system: managers get metrics, admins get controls, developers get PR/debugging workflow improvements, and power users get more model-key flexibility in VS Code.
Related April coverage: AI Industry Roundup, April 24 captured the GPT-5.5-in-Copilot signal; GitHub Copilot adds GPT-5.5 is the direct rollout note; AI News Desk, April 25 folded that into the weekend GPT-5.5, Copilot, Project Deal, and Google-Anthropic catchup.
May 1 security coverage added a protocol-level caveat: the MCP STDIO flaw turns default agent connectors into shell-access surfaces. Copilot teams using MCP should inventory configs, disable auto-registration where possible, sandbox tools, and treat every STDIO server as privileged execution.
April 30 security coverage added another practical caveat: recent coding-agent exploits keep targeting credentials, not model weights. Copilot’s Coding Agent should be governed like CI infrastructure: least privilege, separated environments, secrets hygiene, and auditable action logs.
System Verdict
Pick GitHub Copilot if you’re already inside the GitHub ecosystem and want AI inside your current IDE at the lowest credible price. Pro at $10/mo is the cheapest entry point in serious AI coding. The paid model picker spans OpenAI, Anthropic, Google, and xAI options across Copilot surfaces, but model availability is now explicitly surface-specific after GitHub removed Gemini models from Copilot Chat on GitHub.com on May 20, 2026.
The autonomous Coding Agent accepts a GitHub issue, spins up a cloud dev environment via Actions, self-reviews the patch, and opens a PR. No direct equivalent in Cursor or Claude Code. IDE coverage is unmatched: JetBrains, Xcode, and Neovim developers have no Cursor path.
Skip it if you want a pure autonomous terminal loop or a GUI-first multi-agent workbench. Claude Code runs a cleaner CLI agent loop on test failures and build errors. Cursor 3.0 ships an Agents Window with parallel worktree and cloud agents that Copilot doesn’t match.
Heavy Opus 4.7 users on Pro+ will burn the 1,500-request pool fast. At the 7.5x promotional multiplier, that’s roughly 200 Opus 4.7 turns per month.
Who pays which tier: Free for evaluation and students, Pro $10/mo for most individuals, Pro+ $39/mo for power users who need the full model picker and Spark access, Business $19/seat for teams wanting IP indemnification and policy controls, Enterprise $39/seat for org-scale knowledge features and SSO/audit. Enterprise also requires GitHub Enterprise Cloud at $21/user/mo on top.
Key Facts
| Flagship models | GPT-5.5 (GA April 24, 2026) · Claude Opus 4.7 (GA April 16, 2026) on Pro+, Business, Enterprise |
| Other models in picker | GPT-5.3-Codex · GPT-5-Codex · Claude Sonnet 4.6 · Gemini 3.5 Flash and other Gemini access where supported by surface · Grok · Claude Haiku 4.5 |
| Free-tier models | Claude Sonnet 4.6 · GPT-4.1 only |
| Modes | Ask · Edit · Agent · Coding Agent (autonomous, cloud) · Code Review · Copilot CLI · PR chat/debugging workflows |
| IDE support | VS Code · Visual Studio · JetBrains · Xcode · Neovim (chat limited to first three) |
| Agent mode status | GA on VS Code and JetBrains since March 2026; MCP tool use supported |
| Coding Agent status | GA for all paid Copilot subscribers; runs in GitHub Actions sandbox, self-reviews, runs security scans |
| Subscription pricing | Free · Pro $10 · Pro+ $39 · Business $19/seat · Enterprise $39/seat |
| Premium request caps | Free 50 · Pro 300 · Pro+ 1,500 · Business 300/seat · Enterprise 1,000/seat · overage $0.04 each |
| Copilot Spaces | Context-scoping feature accessible via the GitHub MCP server |
| Data-training policy | Free / Pro / Pro+ interaction data may be used for model training unless opted out; Business / Enterprise excluded |
| Recent shipments (12 mo) | Semantic issue search · Auto model selection in VS Code · Copilot for Eclipse open-sourced · GitHub-owned report URLs · Copilot Chat on GitHub.com model cleanup · Gemini 3.5 Flash GA · remote Copilot CLI control GA · Copilot Spaces API GA · repository cloud-agent configuration audit API · cheaper cloud-agent models · Copilot App technical preview · Agents secrets/variables · code-review metrics by comment type · Rubber Duck cross-model review · GPT-5.5 GA (Apr 24, 2026) · BYOK in VS Code · cloud-agent metrics · Jira controls · PR chat/debugging improvements · Coding Agent GA · Agent mode GA (Mar 2026) · MCP support GA · Copilot CLI GA (Apr 2026) · Opus 4.7 GA (Apr 16, 2026) · Agentic code review (Mar 2026) |
What it actually is
A first-party extension bundle covering autocomplete, inline edits, chat, agentic multi-file work, and an autonomous GitHub-native agent that runs in the cloud. One subscription buys access to a curated model picker (Opus 4.7, GPT-5.5, GPT-5.3-Codex, Gemini 3.5 Flash, Grok, Sonnet 4.6) billed via a single premium-request pool. No per-model setup. No BYO API keys.
Three moats, none easy to replicate.
First, GitHub-native integration. Coding Agent assigns to a GitHub issue like a teammate, runs in an Actions-backed sandbox, and opens a PR against the right branch. No third-party tool is wired this deep into the repo, issue, and PR graph.
Second, enterprise distribution. Copilot is already a line item in every Microsoft/GitHub enterprise agreement. IP indemnification on Business and Enterprise is a lawyer-tested feature Cursor and Claude Code don’t match.
Third, IDE breadth. JetBrains, Xcode, and Neovim developers only have Copilot among first-tier AI coding tools with OAuth-simple setup. VS Code now also matters as a control plane for advanced users: BYOK, web debugging, PR chat, and cloud-agent visibility all reinforce Copilot as a workflow layer around the repo, not just an inline completion engine.
When to pick GitHub Copilot
- You already pay for GitHub. Copilot inherits your OAuth, your SSO, and your repo permissions. Setup is a VS Code extension install.
- You use JetBrains, Xcode, or Neovim. Cursor is a VS Code fork with no path to those editors. Copilot is the only first-tier AI coding tool shipping native extensions for all of them.
- You want issue-to-PR automation with governance. Coding Agent turns an assigned GitHub issue into a self-reviewed PR autonomously, while Jira controls and cloud-agent metrics make it easier for teams to supervise work in flight. No equivalent exists in Cursor or Claude Code.
- You want $10/mo entry pricing with a real model picker. Pro at $10 includes a curated model picker, Gemini access, and agent mode; GPT-5.5 and Opus 4.7 require Pro+ or team tiers. Cheapest serious AI coding subscription on the market.
- You need IP indemnification. Business and Enterprise tiers include Microsoft-backed IP indemnity for Copilot suggestions. No direct analog from Cursor or Anthropic.
- You want MCP tool use without leaving the IDE. Agent mode autonomously invokes MCP servers once configured, and the GitHub MCP server now exposes Copilot Spaces for scoped context.
When to pick something else
- Pure CLI autonomous agent loop: Claude Code. Stronger terminal-native loop on test failures and build errors; runs the whole cycle without IDE supervision.
- GUI-first multi-agent workbench: Cursor. Cursor 3.0’s Agents Window orchestrates parallel agents across local worktrees, cloud sandboxes, and remote SSH with Design Mode for UI clicks.
- Open-source agent inside stock VS Code: Cline. BYO API key, no editor fork, no premium-request pool.
- Terminal pair-programmer with precise diff control: Aider. Git-native, surgical edits, popular with power users.
- Cursor-style editor on a tighter budget: Windsurf. Similar ergonomics at a lower sticker price.
- Fully configurable, self-hostable VS Code extension: Continue. Bring your own models and policies.
Pricing
Subscription tiers via github.com/features/copilot/plans and docs.github.com/copilot/get-started/plans:
| Plan | Price | Premium requests | Model access | Who’s it for |
|---|---|---|---|---|
| Free | $0 | 50/mo | Sonnet 4.6 · GPT-4.1 · 2,000 completions | Students, evaluation, OSS maintainers |
| Pro | $10/mo | 300/mo | Full picker (Opus 4.7 not included; see note) · agent mode · unlimited completions | Most individuals should land here |
| Pro+ | $39/mo | 1,500/mo | Full picker including GPT-5.5 and Opus 4.7 · GitHub Spark · early access | AI power users burning Pro limits weekly |
| Business | $19/user/mo | 300/user/mo | Full picker including GPT-5.5 and Opus 4.7 · org policy · IP indemnity · audit | 2+ seat teams on GitHub Enterprise |
| Enterprise | $39/user/mo | 1,000/user/mo | Full picker including GPT-5.5 and Opus 4.7 · org knowledge · SSO · SCIM · audit logs | Compliance-heavy orgs (also needs GitHub Enterprise Cloud at $21/user/mo) |
Overage: $0.04 per premium request on every paid tier. Opus 4.7 is billed from the same pool at a 7.5x promotional multiplier through April 30, 2026. On Pro+‘s 1,500-request pool, that works out to roughly 200 Opus 4.7 turns per month. The post-promotion multiplier has not been announced.
Prices verified 2026-04-17 via GitHub Copilot plans, docs.github.com individual plans, the Opus 4.7 GA changelog, and the Opus 4.6 Fast retirement changelog.
Against the alternatives
| GitHub Copilot Pro $10 | Cursor Pro $20 | Claude Code (via Claude Max 5x $100) | |
|---|---|---|---|
| IDE integration | Native extension in VS Code, Visual Studio, JetBrains, Xcode, Neovim | VS Code fork only | Terminal-first, no IDE |
| Model access | Picker: GPT-5.5 and Opus 4.7 (Pro+ and above) · GPT-5-Codex · Gemini 3.1 Pro · Grok · Sonnet 4.6 | Picker: Opus 4.7 · GPT-5.5 · Gemini 3.1 Pro · Grok · Kimi · Composer 2 | Opus 4.7 · Sonnet 4.6 · Haiku 4.5 only |
| Agent quality | Agent mode (IDE) + Coding Agent (async, GitHub-native) | Agents Window + Cloud Agents (supervised workbench) | Strongest autonomous CLI loop |
| GitHub-native features | Issue-to-PR via Coding Agent · PR review agent · Actions integration · Spaces · IP indemnity | None beyond a repo connection | None native |
| Pricing | $10/mo flat + 300 premium requests (Opus locked to Pro+ at 7.5x) | $20/mo + $20 usage pool at API rates | $100/mo flat for Max 5x |
| Best viewed as | GitHub-native AI inside your existing IDE | GUI-first multi-agent workbench | Strongest autonomous CLI agent |
Recent changes
- May 20-21, 2026: GitHub Copilot added semantic issue search, auto model routing, Eclipse transparency, GitHub-owned report URLs, and web model cleanup. Teams should treat model availability as surface-specific, especially because Gemini models were removed from Copilot Chat on GitHub.com on May 20.
- May 19, 2026: Gemini 3.5 Flash became generally available in GitHub Copilot, reinforcing Copilot as a multi-model coding surface where teams need model governance instead of one-provider assumptions.
- May 18, 2026: GitHub turned Copilot into a more governable coding-agent control plane with remote CLI control GA, Copilot Spaces API GA, cheaper cloud-agent models, and repository cloud-agent configuration audit APIs.
- May 11, 2026: OpenAI launched Daybreak, pulling Codex Security into a 22-partner cyber initiative. Copilot teams now have a productized OpenAI security-scanning surface that competes with Anthropic’s Project Glasswing and Claude Security.
- May 8, 2026: GitHub will deprecate Grok Code Fast 1 across Copilot on May 15. Teams using it should switch model policies to GPT-5 mini, Claude Haiku 4.5, or another supported model before the deadline.
- May 8, 2026: Copilot cloud agent gained dedicated Agents secrets and variables, including organization-level sharing for agent configuration.
- May 8, 2026: Copilot code-review metrics now include comment-type breakdowns, giving admins a better view of security, bug-risk, and other suggestion categories.
- May 7, 2026: Copilot CLI Rubber Duck can now pair GPT and Claude as cross-model critics when experimental mode is enabled.
- May 7, 2026: GitHub retired Claude Sonnet 4 and set GPT-4.1’s Copilot retirement for June 1. Enterprise admins should verify GPT-5.5 and Claude Sonnet 4.6 are enabled where needed.
Failure modes
- Model retirements are accelerating. In May 2026 alone, GitHub retired Claude Sonnet 4, set GPT-4.1 for June 1, and set Grok Code Fast 1 for May 15. Copilot buyers need a model-policy owner and a fallback testing process.
- Model availability is surface-specific. Gemini 3.5 Flash can be part of Copilot while all Gemini models are removed from Copilot Chat on GitHub.com. Check the exact surface: VS Code, GitHub.com web chat, CLI, Eclipse, JetBrains, cloud agent, Business, and Enterprise do not always expose the same model list.
- Model churn is now operational. GPT-5.2 and GPT-5.2-Codex retire from most Copilot experiences on June 1, 2026. Admins need model-policy owners, not just developer preference lists.
- Premium request quota burns fast on GPT-5.5 and Opus 4.7. At the 7.5x multiplier, Pro+‘s 1,500 requests yield ~200 top-model turns. Heavy users should pin Sonnet 4.6 or GPT-5-Codex for routine work and reserve GPT-5.5 / Opus for hard problems, or plan for overage at $0.04/request.
- Opus 4.7 is not in Pro ($10). Access requires Pro+ ($39), Business, or Enterprise. Pro is oriented around lower-cost coding and Gemini/Sonnet models by default.
- Post-April 30 Opus 4.7 multiplier is unannounced. Budget accordingly: the 7.5x figure is explicitly “promotional.”
- Autocomplete latency varies by language. Strongest on Python, TypeScript, JavaScript, Go. Noticeably weaker than dedicated IDE stacks on niche languages and on large cold files.
- GitHub lock-in is real. Coding Agent, PR review, Spaces, and org knowledge all assume GitHub is your forge. GitLab and Bitbucket teams get the extension but lose the agent graph.
- Enterprise admin surface is complex. Policy controls, per-model allow/deny lists, content exclusions, and SKU combinations (Copilot Enterprise + GitHub Enterprise Cloud) take real time to configure correctly.
- Auth quirks on multi-account setups. Users with multiple GitHub accounts (personal plus corporate) regularly hit sign-in confusion across the VS Code, JetBrains, and Copilot CLI surfaces; the fix is usually signing out of all GitHub integrations and signing back in on the intended account.
- Free / Pro / Pro+ data-training opt-out is now a real buying criterion. As of April 24, 2026, individual-tier interaction data can feed model training unless users opt out. Teams with sensitive code should prefer Business or Enterprise policy controls.
- Pro+ rate limits exist beyond the 1,500-request quota. Community reports of
user_weekly_rate_limitederrors suggest undocumented per-week caps; heavy users should monitor for this. - Coding Agent needs well-scoped issues. Vague tickets produce bad PRs. Works best when the issue has acceptance criteria, file hints, and a reproduction: essentially the same hygiene that makes a junior engineer productive.
Methodology
This page was produced by the aipedia.wiki editorial pipeline, an automated system that ingests vendor documentation, normalizes factual claims, verifies pricing and model details against primary sources, and generates the editorial analysis you are reading. No individual human wrote this review. Scoring follows the four-dimension rubric at /about/scoring/ (Utility × Value × Moat × Longevity, unweighted average). Last verified 2026-05-22 against github.com/features/copilot, github.com/features/copilot/plans, docs.github.com/en/copilot, the May 20-21 semantic search / auto model / Eclipse / web model update, the May 19 Gemini 3.5 Flash GA update, the May 18 Copilot agent control-plane update, the May 11 OpenAI Daybreak launch, the May 8 Grok Code Fast 1 retirement, the May 8 cloud-agent secrets update, the May 8 code-review metrics update, the May 7 Rubber Duck update, and the May 7 model retirement notices.
FAQ
Is GitHub Copilot free? Yes. The Free tier gives 2,000 code completions and 50 premium requests per month on Claude Sonnet 4.6 and GPT-4.1. Students, teachers, and maintainers of popular open-source projects get Copilot Pro at no cost. Agent mode and the full model picker require Pro or above.
Does GitHub Copilot include GPT-5.5? Yes, on Pro+, Business, and Enterprise. GPT-5.5 became generally available inside Copilot on April 24, 2026 and uses premium requests at a 7.5x multiplier. Pro ($10) does not include GPT-5.5.
Does GitHub Copilot include Claude Opus 4.7? Yes, but not on every tier. Opus 4.7 went GA inside Copilot on April 16, 2026 for Pro+, Business, and Enterprise. It is billed from the premium-request pool at a 7.5x multiplier through April 30, 2026. Pro ($10) does not include Opus 4.7. Upgrade to Pro+ ($39) or Business/Enterprise for access.
What’s the difference between Agent mode and the Coding Agent? Agent mode runs inside your IDE (VS Code, JetBrains, Visual Studio) and edits your open repo interactively: you watch it think, it asks for approval on sensitive actions, and iterates on errors. The Coding Agent is asynchronous and cloud-hosted: you assign a GitHub issue to Copilot, it spins up a GitHub Actions sandbox, writes and self-reviews code, runs security scans, and opens a PR for your review. Agent mode is for active coding; Coding Agent is for backlog triage.
How does Copilot compare to Cursor? Copilot is a first-party extension to your existing IDE (VS Code, JetBrains, Visual Studio, Xcode, Neovim); Cursor is a VS Code fork. Copilot Pro is $10/mo vs Cursor Pro $20/mo + usage pool. Cursor’s Agents Window is a stronger supervised multi-agent workbench; Copilot’s Coding Agent is stronger on GitHub-native issue-to-PR automation. JetBrains, Xcode, and Neovim developers can’t use Cursor; Copilot is the only option there.
What IDEs does Copilot support? VS Code, Visual Studio, JetBrains IDEs (IntelliJ, PyCharm, WebStorm, Rider, GoLand, etc.), Xcode, and Vim/Neovim. Chat functionality is available in VS Code, JetBrains, and Visual Studio; autocomplete is available everywhere. Edit mode is VS Code and JetBrains. Agent mode is VS Code and JetBrains GA (feature matrix).
Does Copilot support MCP? Yes. Agent mode in VS Code and the Coding Agent both support the Model Context Protocol for tool use. Configured MCP servers are invoked autonomously by the agent without per-call approval once authorized (enhance agent mode with MCP). The GitHub MCP server now exposes Copilot Spaces for scoping a task to a curated bundle of code, docs, and issues.
Related
- Category: AI Coding
- Alternatives: Cursor · Claude Code · Cline · Aider · Windsurf · Continue
- Comparisons: Cursor vs Claude Code vs GitHub Copilot · Cursor vs GitHub Copilot · Claude Code vs GitHub Copilot · Continue vs GitHub Copilot
