GitHub Copilot review: still worth the $20/month?
GitHub Copilot review: still worth the $20/month?
Yes, for most working developers, GitHub Copilot is still worth $20 a month in 2026. But it is no longer the best AI coding tool, and for anyone doing agentic, multi-file, or refactor-heavy work, Cursor or Claude Code will get more done per dollar.
That is the short answer. The long answer involves Copilot Workspace, the new agent mode, the GPT-5 / Claude 4.5 model picker, and a frank look at where Copilot still wins, where it has fallen behind, and who should pick something else.
What GitHub Copilot actually is in 2026
Copilot is now three products in a trench coat.
The original product, inline completion, still ships as the editor extension that suggests grey-text completions as you type in VS Code, JetBrains, Neovim, Visual Studio, and Xcode. This is what most people picture when they hear "Copilot."
The second product is Copilot Chat, an in-editor chat panel that takes your selected file or repo context and answers questions, generates patches, and writes tests. It now supports a model picker (GPT-5, GPT-5 mini, Claude Sonnet 4.5, Claude Opus 4.5, Gemini 2.5 Pro, o4) on the Pro and Business tiers.
The third product is Copilot Coding Agent and Copilot Workspace, GitHub's answer to Cursor's agent mode and Devin. You assign an issue to Copilot, it spins up a sandboxed environment, drafts a PR, and pings you for review. This is the part GitHub has been pushing hardest in 2026, and it is also the part with the most mixed results.
What it is not: a full IDE replacement. The autocomplete and chat features still live inside whatever editor you already use. If you want a Copilot-first IDE, you switch to Cursor or Windsurf.
Pricing in 2026
| Plan | Price | Who it is for | Model picker | Agent minutes |
|---|---|---|---|---|
| Copilot Free | $0 | Trying it out | GPT-5 mini only | 50/mo |
| Copilot Pro | $10/mo | Solo devs, students | Full picker | 300/mo |
| Copilot Pro+ | $39/mo | Heavy individual use | Full picker + Opus 4.5 | 1,500/mo |
| Copilot Business | $19/user/mo | Teams | Full picker | Unlimited completions |
| Copilot Enterprise | $39/user/mo | Larger orgs | Full + custom models | Unlimited + audit |
The $20/month figure in this post's title is an approximation of the Business plan ($19) that most readers actually pay, or the older $19 individual plan that has now been split into Pro and Pro+.
A few price gotchas worth flagging:
- Premium request limits. Pro caps you at 300 "premium" requests per month (anything that hits Claude Opus, GPT-5, or the agent). After that you fall back to GPT-5 mini, which is fine for completion and weak for chat reasoning.
- Agent minutes are billed. Copilot Coding Agent burns minutes the same way GitHub Actions does. A junior dev assigning Copilot to ten issues a day will exhaust the Pro+ allotment in two weeks.
- No bring-your-own-key. Unlike Cursor, you cannot point Copilot at your own Anthropic or OpenAI API key. You pay GitHub's markup or you do not use the model.
Where Copilot still wins
1. Boilerplate and inline completion in mainstream stacks
For TypeScript, Python, Go, Java, and C#, Copilot's inline suggestions remain the fastest and least-disruptive of any tool we tested. The latency is sub-200ms in most regions. The accept rate on suggestions over 5 lines holds around 35-40% for typical app code, which matches what most published studies report.
Cursor's tab completion is now competitive (sometimes better at multi-line edits inside a file), but Copilot's single-line completion still feels the most like a fast keyboard. If you mostly write CRUD endpoints, React components, and standard test files, this matters.
2. The IDE you already use
Copilot ships as a first-class extension in VS Code, JetBrains (IntelliJ, PyCharm, GoLand, WebStorm, Rider), Neovim, Vim, Visual Studio, Xcode, and Eclipse. Cursor and Windsurf are forks of VS Code; if your team uses Rider or PyCharm, switching to them means rebuilding muscle memory and giving up paid JetBrains plugins.
For shops that have standardized on JetBrains, Copilot is the only credible option without a wholesale IDE migration.
3. GitHub-native context
Copilot Chat reads issues, PRs, discussions, and Actions logs from your repo without any setup. Ask "why did the deploy fail last night" and it grabs the failed Actions run and explains it. Cursor and Claude Code can do similar things via MCP servers, but Copilot ships it for free out of the box.
For teams already living in GitHub (so, most teams), this saves 15-20 minutes a day of context-switching.
4. Compliance, SSO, and audit logs
Copilot Business and Enterprise have the SOC 2 Type II reports, the SSO/SCIM hooks, the IP indemnification, and the audit log surface that procurement teams want. Cursor has been catching up here through 2026, but Copilot is still the safer pick for a large enterprise rollout.
If your security team has a 47-question vendor questionnaire, Copilot has answers to all of them.
Where Copilot has fallen behind
1. Multi-file refactors
This is the clearest gap. Ask Cursor's Composer to "rename User.email to User.contactEmail across the codebase, update the migration, and fix all usages including tests" and it does it in one pass with an editable diff. Ask Copilot Chat the same thing and you get partial edits, missed test files, and a broken migration about half the time.
Copilot Workspace tries to solve this with a "spec → plan → patch" agent flow, but in practice the plans are over-engineered and the patches still miss files outside the obvious search radius.
2. Agentic, long-horizon tasks
Copilot Coding Agent works for the canonical demo (small bug, well-isolated module, good test coverage). On a real codebase with 80k lines and partial test coverage, it routinely:
- Misreads the test setup and submits PRs that fail CI
- Picks the wrong abstraction layer to modify
- Takes 25+ minutes per task, because it spawns a fresh sandbox for every run
Claude Code, used locally with the same repo, finishes the same tasks in 4-6 minutes because it keeps state and runs the test suite incrementally. Cursor's Background Agents are also faster because they run inside your existing dev container.
The agent pricing also bites. A senior engineer assigning ten tickets a day hits the Pro+ limit fast, and Business does not currently include unlimited agent minutes.
3. The model picker quality gap
Copilot lets you pick Claude Sonnet 4.5 or Claude Opus 4.5, but the way it injects context, the system prompt, and the tool definitions is tuned for OpenAI models. The same model performs noticeably better in Cursor or Claude Code on the same task. This is not a benchmark we can publish; it is the consistent feedback from senior engineers who have used both. If you are paying for Opus, you want the harness Anthropic actually optimizes for.
4. Spec-driven and prompt-as-spec workflows
The teams getting the most out of AI in 2026 are writing structured specs (docs/specs/feature-x.md), pasting them into the agent, and reviewing the diff. Cursor has Rules and Composer to support this. Claude Code has CLAUDE.md and explicit subagents. Copilot's "custom instructions" feature is shallower (one global file, no per-task overrides) and the agent ignores them about a third of the time. If you have read our take on what AI-native engineering actually means, this is the gap that hurts most.
Copilot vs the alternatives
| Tool | Best at | Worst at | Price |
|---|---|---|---|
| GitHub Copilot | Inline completion, JetBrains IDE, GitHub-native context, enterprise compliance | Multi-file refactors, agentic tasks, prompt-as-spec workflows | $10-39/user/mo |
| Cursor | Multi-file edits, Composer, tab completion, Background Agents | Locked into VS Code fork, weaker GitHub integration | $20-60/user/mo |
| Claude Code | Long-horizon agent tasks, terminal-native workflows, CLAUDE.md context | No GUI, steep ramp for non-CLI devs | $20/mo (Pro) or API-billed |
| Sourcegraph Cody | Massive monorepo context, code search, enterprise scale | Slower inline completion, smaller community | $9-19/user/mo |
| Continue (OSS) | Self-hosted, BYOK, customizable | You own the integration work | Free + your API costs |
The honest take: in 2026 there is no single "best" tool. Most senior engineers we work with run two of these in parallel. A common combo is Copilot for inline completion in their IDE, plus Claude Code in a terminal for longer agent tasks. That stack costs about $30/month and outperforms either tool used alone.
Who should buy Copilot in 2026
Pick Copilot Pro ($10/mo) if:
- You write code daily in mainstream languages
- You use VS Code or JetBrains and do not want to switch IDEs
- You mostly want inline completion, not agentic refactors
- You already live in GitHub and want chat over your repo
Pick Copilot Business ($19/user/mo) if:
- You are rolling out AI tooling to a team of 5+ engineers
- You need SSO, audit logs, and IP indemnification
- Your team is heterogenous (mix of VS Code and JetBrains users)
Pick Copilot Pro+ ($39/mo) if:
- You actively use the Coding Agent for real work and are hitting limits
- You want priority access to Opus 4.5 and GPT-5
Skip Copilot and pick something else if:
- You do heavy multi-file refactors → Cursor
- You want a strong agentic flow → Claude Code
- You work in a 1M+ LOC monorepo → Sourcegraph Cody
- You want to BYOK and self-host → Continue or Aider
What to do this week
If you are already paying for Copilot, do not cancel reflexively. Run a one-week experiment: keep Copilot for completion, add Claude Code's $20/mo Pro plan, and use Claude for any task that touches more than two files. Track whether your PR throughput moves. If it does, you have your stack. If it does not, drop Claude and stay on Copilot alone.
If you are evaluating from scratch, the cheapest honest test is the Copilot 30-day free trial plus the Cursor 14-day free trial in parallel. Use each for half your work for a week. The one you reach for unprompted on day five is your tool.
If you are a founder hiring engineers and wondering whether to filter for "AI-native" candidates: don't. The discriminator is no longer whether someone uses Copilot. It is whether they have a working point of view on which task goes to which tool. Every engineer on Cadence is AI-native by default, vetted on Cursor, Claude Code, and Copilot fluency in a voice interview before they unlock bookings, so you skip that filtering step entirely. You can also read our take on what AI-native actually means before you write the next job spec.
If you want a quick read on whether your current dev tooling is pulling its weight, our Ship-or-Skip tool audit takes about 4 minutes and grades your stack honestly. It will not try to sell you Copilot.
The Cadence stack note
Every engineer on the Cadence platform uses some combination of Cursor, Claude Code, and Copilot daily. We do not mandate a specific tool because the right pick varies by task: Copilot for IDE-native completion in JetBrains, Cursor for big refactors, Claude Code for agent runs. The voice interview tests whether the candidate can articulate which tool they reach for and why. We currently match against a pool of 12,800 engineers in 80ms when a founder books, and the median time to first commit is 27 hours.
If you want to compare AI assistants on the model side, our ChatGPT vs Claude breakdown for developers goes deeper on which underlying model handles which kind of task best. That comparison matters more than the editor wrapper in 2026.
Tooling decisions are reversible. Hiring decisions are slower. If you would rather book an engineer who already runs this stack than spend three weeks vetting AI fluency in interviews, browse Cadence and try someone for 48 hours free.
FAQ
Is GitHub Copilot worth it in 2026?
For most working developers writing code in mainstream languages, yes. Copilot Pro at $10/month or Business at $19/user/month pays back the cost in saved typing alone, especially for inline completion. Skip it only if you do mostly multi-file refactors or agentic work, in which case Cursor or Claude Code is a better primary tool.
GitHub Copilot vs Cursor: which should I pick?
Pick Cursor if multi-file edits and the Composer flow matter most, or if you are happy on a VS Code fork. Pick Copilot if you use JetBrains, want first-class GitHub integration, or need enterprise compliance and audit logs. Many senior engineers run both: Copilot for inline completion, Cursor for refactors.
Can I use GitHub Copilot for free?
Yes. Copilot Free gives you GPT-5 mini completions, 50 chat messages a month, and 50 agent minutes a month. It is enough to test the product but not enough to use it as a daily driver. Students and verified open-source maintainers also get full Copilot Pro free through GitHub Education.
How does Copilot compare to Claude Code?
They are different categories. Copilot lives inside your IDE and is best at inline completion and quick chat. Claude Code is a terminal-native agent that is better at long-horizon multi-file tasks and projects with rich CLAUDE.md context. Most senior engineers use both: Copilot in the editor, Claude Code in a terminal pane.
Does Copilot store or train on my code?
On Pro and Business, GitHub commits to not training models on your private code, and on Business you can disable telemetry entirely. Public-repo code may be used per the standard GitHub terms. If your security team needs zero-data-retention guarantees, Copilot Enterprise is the tier that ships them in writing.
Is the Copilot Coding Agent ready for production work?
For small, well-scoped issues with strong test coverage, yes. For larger refactors or anything ambiguous, it still misfires often enough that you should review every PR carefully. As of mid-2026 we would not let it merge unattended, and the per-task minute cost adds up fast for high-volume use.