Take Microsoft GH-300 if you need to prove you can use and administer GitHub Copilot, including agent mode, as a developer or team lead. Take the Microsoft GH-600 if you configure, supervise, and govern AI agents inside CI/CD and production workflows. Neither exam is a formal prerequisite for the other, but GH-600 assumes you already work comfortably with Copilot, MCP servers, and custom agents. Here is a detailed comparison between these two Microsoft certifications.
| Verdict | GH-300: GitHub Copilot | GH-600: Agentic AI Developer |
| Difficulty (our rating) | 3/5, moderate | 4/5, moderately hard |
| Published pass rate | None. Microsoft and GitHub do not publish pass rates for either exam, so treat any percentage on prep sites as unverified | None |
| Prep reality (our estimate) | 15 to 30 hours for daily Copilot users | 40 to 60 hours, including hands-on agent labs |
| Best for | Developers, Copilot admins, technology managers | Platform, DevOps, AI, and security engineers running agents |
Specs verified against the official Microsoft Learn certification pages and study guides in September 2026. Difficulty and hour estimates are editorial judgments based on each exam’s published scope, not official figures.
What is the difference between GH-300 and GH-600?
Microsoft GH-300 tests whether you can use GitHub Copilot productively and responsibly. Microsoft GH-600, on the other hand, tests whether you can operate and control AI agents that act on their own inside the software development lifecycle (SDLC).
GH-300 (GitHub Copilot)
- Focus: Copilot in the IDE and CLI, Copilot Chat, agent mode, prompt engineering, responsible AI, organization policies, and privacy controls.
- Audience: Developers, DevOps engineers, app makers, and technology managers who use or roll out Copilot.
- Status: Generally available. The current skills outline took effect August 7, 2026.
GH-600 (GitHub Certified: Agentic AI Developer)
- Focus: Agent architecture, tool use and MCP servers, memory and state, evaluation and tuning, multi-agent orchestration, and guardrails, using GitHub as the system of record and control plane.
- Audience: AI engineers, solution architects, DevOps and platform engineers, data engineers, and developers who deploy agents in production workflows.
- Status: Generally available. The beta label was removed in late July 2026 (details below).
Microsoft GH-300 vs GH-600: How do the exam specs compare?
| Spec | GH-300 | GH-600 |
| Official page | GitHub Copilot certification | GitHub Certified: Agentic AI Developer |
| Level | Intermediate | Intermediate |
| Duration | 100 minutes | 120 minutes |
| Languages | English, Spanish, Portuguese (Brazil), Korean, Japanese | English only |
| Passing score | 700 | 700 |
| Question count | Not published | Not published |
| Question formats | Proctored; may include interactive components | Proctored; may include interactive components |
| Official practice assessment | Yes, free | Not offered on the certification page |
| Exam sandbox | Yes | Yes |
| Microsoft Learn role tags | App Maker, Developer, DevOps Engineer, Technology Manager | AI Engineer, App Maker, Data Engineer, Developer, DevOps Engineer, Solution Architect |
| Domains | 6 | 6 |
| Scheduling | Pearson VUE | Pearson VUE |
| Validity | 2 years | 2 years |
GitHub states its certifications are valid for 2 years, and certifications expiring before GitHub moves to Microsoft’s recertification process get a 6-month extension. The study guides’ generic note that Microsoft certifications “expire annually” refers to Microsoft’s own associate and expert credentials, not GitHub certifications.
Read more: How to Prepare for GH 600 in 40 Hours: Which Domains Matter Most?
Is GitHub GH-600 still in beta?
No. GH-600 launched in beta in May 2026 and became generally available in late July 2026, when Microsoft removed “beta” from the certification title. Pages and AI summaries that still call it a beta exam are out of date.
- Beta discount: The first 100 candidates who sat the exam by May 31, 2026, received 80% off. That offer has ended.
- Beta results: Beta scores are released about 10 days after an exam goes live.
- Same credential: Passing the beta or the live exam earns the same certification.
Where do GH-300 and GH-600 overlap?
Both exams now cover agent mode, MCP, and responsible AI, which is why many summaries blur them. The difference is depth: GH-300 asks you to use a feature, while GH-600 asks you to configure, restrict, and audit it.
| Topic | GH-300 tests you on | GH-600 tests you on |
| Agent mode | Triggering agent mode, managing agent sessions, delegating to sub-agents | Separating planning from execution, validating agent plans, blocking action until approved |
| MCP | Using MCP to enhance development workflows | Adding MCP servers as tools, configuring GitHub remote MCP servers, MCP registries, and MCP allow lists |
| Customization | Instruction files for review standards, reusable prompt files | Configuring custom agents, their tools, and tool permissions |
| Responsible AI | Risks, limitations, and validating AI output | Blocking actions that violate Responsible AI policies, assigning autonomy levels by risk |
| Security and privacy | Content exclusions, public code filtering | Least-privilege scopes, human approval for irreversible or compliance-sensitive changes |
| Code quality | Generating unit and integration tests, finding edge cases | Generating evaluation signals with automated scanning, classifying root causes of agent failures |
| CI/CD | Not a core objective | Invoking agents in CI workflows, branch-based scope, agents creating branches and pull requests |
| Context and memory | How Copilot determines context, chat history | Short-term, long-term, and external memory, context drift, persisting state across tools |
| Multiple agents | Delegating tasks to sub-agents | Orchestration patterns, agent isolation, resolving conflicting outputs, retiring agents with auditability |
| Governance | Organization policies, audit log events, subscriptions via REST API | Traceability of agent actions, audit-ready artifacts, rollback and escalation paths |
Why is content exclusion the dividing line?
Content exclusion is the main GH-300 control for keeping files away from Copilot, but GitHub documents that Copilot CLI, Copilot coding agent, and agent mode in Copilot Chat in IDEs do not support it. Once Copilot acts as an agent, you need the controls GH-600 covers: scoped tool permissions, MCP allow lists, repository and branch scoping, and human-in-the-loop gates.
How are the Microsoft GH-300 and GH-600 exam domains weighted?
Here is a look at the Microsoft GH-600 official domains:
| Domain | Official weight | Suggested share of study time* |
| Prepare agent architecture and SDLC processes | 15-20% | 17.5% |
| Implement tool use and environment interaction | 20-25% | 22.5% |
| Manage memory, state, and execution | 10-15% | 12.5% |
| Perform evaluation, error analysis, and tuning | 15-20% | 17.5% |
| Orchestrate multi-agent coordination | 15-20% | 17.5% |
| Implement guardrails and accountability | 10-15% | 12.5% |
Here is a look at the Microsoft GH-300 official domains:
| Domain | Official weight | Suggested share of study time* |
| Use GitHub Copilot responsibly | 15-20% | 18% |
| Use GitHub Copilot features | 25-30% | 29% |
| Understand GitHub Copilot data and architecture | 10-15% | 13% |
| Apply prompt engineering and context crafting | 10-15% | 13% |
| Improve developer productivity with GitHub Copilot | 10-15% | 13% |
| Configure privacy, content exclusions, and safeguards | 10-15% | 13% |
*Our calculation: the midpoint of each official range, normalized to 100%.
Which is harder, GH-300 or GH-600?
GH-600 is harder for most candidates. It tests configuration and judgment in scenarios that most Copilot users have not run in production.
What makes GH-300 challenging
- Breadth: IDE, CLI, Chat, agent mode, Spaces, Spark, code review, and admin settings in one exam.
- Admin details that individual users rarely touch, such as organization policies, audit log events, and managing subscriptions through the REST API.
- Blueprint churn: study material written before 2026 misses agent mode, MCP, and sub-agents.
What makes GH-600 challenging
- Hands-on depth: configuring MCP servers, allow lists, CI invocation, and rollbacks cannot be learned from reading alone.
- Judgment questions: choosing autonomy levels and deciding which actions need human approval without slowing delivery.
- Fewer vetted resources: no official practice assessment, and the exam is newer, so third-party questions are less proven.
- Heaviest domain (tool use and environment interaction) is the least familiar area for IDE-focused developers.
Do you need GH-300 before GH-600?
No. Microsoft GH-600 has no required prerequisite certification. The official audience profile expects experience with the SDLC, GitHub workflows and controls, code review and security practices, and coding agents including GitHub Copilot, MCP servers, custom instructions, custom agents, and Copilot setup steps.
Taking GH-300 first makes sense if you have not yet worked with agent mode, MCP, or Copilot policies. Skip straight to GH-600 if you already run agents outside the IDE.
Which GitHub AI certification should you take?
- You use Copilot mainly for suggestions and chat: Take GH-300.
- You manage Copilot seats, policies, or rollout for a team: Take GH-300. Organization policies and audit logs are GH-300 objectives, not GH-600 objectives.
- You already run Copilot coding agent, custom agents, or MCP servers: Take GH-600 directly.
- You are a platform, DevOps, or security engineer responsible for agent guardrails: Take GH-600.
- You are an AI engineer or solution architect: Take GH-600. These roles are tagged on GH-600 only.
- You want to test in a language other than English: Take GH-300. GH-600 is English only.
- You want both credentials: Take GH-300 first, then GH-600. The responsible AI, prompting, and agent mode knowledge carries over.
How much do Microsoft GH-300 and GH-600 cost?
GitHub’s official credentials page lists GitHub certification exams at $99 USD, with regional pricing that varies by country (GitHub Learn). Several third-party prep sites list GH-600 at $165 USD, but no official page we checked confirms that figure. Check the price at Pearson VUE checkout before you budget.
| Scenario | Cost (USD, before regional adjustment) |
| GH-300 only | $99 |
| GH-600 only | $99 (official GitHub list price) to $165 (third-party reported) |
| Both exams | $198 to $264 |
| Each retake | Full exam fee again |
- Regional pricing: Candidates in Europe, India, Australia, and Canada pay the local Pearson VUE price, which may differ from the USD figure.
- Retake insurance: Microsoft periodically offers Exam Replay, which bundles an exam with a retake.
How should you prepare for GH-300 and GH-600?
| Resource | GH-300 | GH-600 |
| Free official training | GitHub Copilot Fundamentals Part 1 and Part 2 | Foundations of Agentic AI in GitHub, Designing Agent Architecture and SDLC Integration, Tooling, MCP, and Agent Execution Environments |
| Official practice test | Free practice assessment | None; build your own labs |
| Key documentation | Plans and features, content exclusions, prompt engineering | Custom agents, Copilot memory, MCP configuration, coding agent guardrails |
GH-300 prep checklist
- Use every mode on a real repository: inline suggestions, Chat, agent mode, Copilot Edits, and the CLI.
- Configure a content exclusion rule and confirm which modes ignore it.
- Map which features depend on the Copilot plan tier.
- Take the official practice assessment and restudy any domain where you miss questions.
GH-600 lab checklist (built from the official objectives)
- Add a GitHub remote MCP server to an agent and restrict it with an MCP allow list.
- Configure a custom agent with minimum tool permissions, scoped to one repository.
- Invoke an agent from a CI workflow with branch-based scope, and let it open a pull request.
- Make the agent output a structured plan and require approval before it acts.
- Force a failure, then implement a retry, a rollback, and an escalation path.
- Run two agents in parallel on the same repository and resolve an overlapping code change.
- Classify five agent actions by risk and assign each an autonomy level.
Are GH-300 and GH-600 worth it for your career?
Both credentials validate skills on the platform where many teams now run AI coding tools, but neither has official salary data tied to it. For context, the U.S. median annual wage for software developers was $135,980 in May 2025, with employment projected to grow 10% from 2025 to 2035.
- GH-300 value: A low-cost signal that you use Copilot with governance and privacy in mind, useful for developers and team leads in organizations adopting Copilot.
- GH-600 value: A more differentiated signal, since it validates agent governance skills that few engineers can yet demonstrate.
- Renewal: GitHub is moving its certifications to Microsoft’s renewal process, which is designed to let you maintain a certification without retaking the full exam.
Can you take GH-600 in India, Europe, Australia, and Canada?
- Europe, Australia, and Canada: Both exams are scheduled through Pearson VUE with country-based pricing.
- India: The GH-600 beta was not available in Turkey, Pakistan, India, or China. The official pages do not state whether this restriction continues after general availability, so confirm availability in Pearson VUE before you prepare.
- Non-native English speakers: If an exam is not offered in your preferred language, the study guides state you can request an additional 30 minutes. This matters most for GH-600, which is English only.
Microsoft GH-600 or GH-300: What is the final verdict?
Microsoft GH-300 and GH-600 are two steps on one path, not competing credentials. GH-300 certifies that you can use and govern GitHub Copilot as a tool, while GH-600 certifies that you can let AI agents act on your repositories without losing control of security, quality, or accountability. If agents already run in your CI pipelines, GH-600 is the credential that matches your work; if they don’t, GH-300 is the faster, lower-risk place to start. Confirm the current price and regional availability at Pearson VUE, then build your study plan around the heaviest domain of the exam you choose.
Frequently Asked Questions (FAQs)
How soon can I retake GH-300 or GH-600 if I fail?
You can retake either exam 24 hours after the first attempt. Wait periods for later attempts vary under Microsoft’s exam retake policy.
Should I register with my work account?
No. Microsoft recommends registering with a personal Microsoft account (MSA). If you register with a work or school account and leave that organization, your exam records are lost and cannot be recovered.
Who maintains the GH-600 exam, GitHub or Microsoft?
Microsoft delivers the exam, but GitHub maintains the exam and the certification.
Do the exams test preview features?
Most questions cover generally available features. Both study guides state that commonly used preview features may also appear.
