How long is the GH-600 exam and what is the passing score?

You have 120 minutes. The exam is scored on a scaled range and requires 700 or greater to pass. It is proctored through Pearson VUE and offered in English only.

How much does GH-600 cost?

Microsoft does not print a single price. The exam fee is set by the country or region where it is proctored, and is shown when you schedule through Pearson VUE.

How long is the GitHub Agentic AI Developer certification valid?

Two years. GitHub is moving its certifications to Microsoft's recertification process, and certifications that expire before that process is available are extended by six months.

Know if you're actually ready. Take the GitHub GH-600 quiz → get your AI readiness report.Take the free test →

Study guide · GitHub GH-600

GitHub GH-600 Exam Guide: Domains, Weights and Format

Q: What is the GitHub GH-600 exam?

GH-600, Developing in Agentic AI Systems, is the exam for GitHub Certified: Agentic AI Developer. It certifies operating, integrating, supervising and governing AI agents in production software delivery workflows, with GitHub as the system of record and control plane.

Q: What are the GH-600 domains and weights?

Six domains: Implement tool use and environment interaction (20-25%); Prepare agent architecture and SDLC processes (15-20%); Perform evaluation, error analysis, and tuning (15-20%); Orchestrate multi-agent coordination (15-20%); Manage memory, state, and execution (10-15%); Implement guardrails and accountability (10-15%).

GH-600 certifies running AI agents inside real software delivery, with GitHub as the control plane. Every domain with its published weight, the format, and who it is written for.

Examifyr·2026·7 min read

What GH-600 is

Exam GH-600, Developing in Agentic AI Systems, leads to GitHub Certified: Agentic AI Developer. Microsoft describes the role as operating, integrating, supervising and governing AI agents inside production-grade software delivery workflows, with GitHub as the system of record and control plane. It is not an exam about building models or prompting a chatbot. It is about what happens when a coding agent opens pull requests in a real repository: what it is allowed to touch, how its plan is checked before it acts, how its output is evaluated, and who approves the irreversible steps.

Exam format at a glance

GH-600 is an intermediate-level, proctored exam delivered through Pearson VUE, and it may include interactive components as well as multiple choice. Microsoft scores on a scaled range with 700 required to pass, which is not the same as answering 70% of questions correctly.

Exam:        Developing in Agentic AI Systems
Code:        GH-600
Credential:  GitHub Certified: Agentic AI Developer
Level:       Intermediate
Duration:    120 minutes
Languages:   English only
Passing:     700, scaled
Price:       set per country; Microsoft prints no figure
Questions:   not published
Valid for:   2 years (moving to Microsoft's renewal process)
Practice:    no official practice assessment yet

NoteGitHub certifications last two years, not one. GitHub says it is moving them to Microsoft's recertification process, and that certifications expiring before that is ready are extended by six months.

The six domains and their weights

Microsoft publishes ranges rather than fixed percentages. Tool use and environment interaction is the single heaviest domain; three more sit at 15-20% each, and the remaining two at 10-15%.

Implement tool use and environment interaction ... 20-25%
Prepare agent architecture and SDLC processes .... 15-20%
Perform evaluation, error analysis, and tuning ... 15-20%
Orchestrate multi-agent coordination ............. 15-20%
Manage memory, state, and execution .............. 10-15%
Implement guardrails and accountability .......... 10-15%

NoteEvery one of the six is about agents. That is unusual: most AI certifications that mention agents give them one domain among several. Here the whole blueprint is agent operations.

What the heaviest domain actually asks

Tool use and environment interaction covers selecting tools and their permissions, adding MCP servers to an agent (including GitHub's remote MCP server, MCP registries and allow lists), scoping an agent to one repository or branch, invoking an agent from a CI workflow, letting it create branches and pull requests, and the unglamorous half: error handling, retries, rollbacks, escalation paths and traceability for every action an agent takes. If you have only used a coding agent interactively in an editor, this is the domain that will feel least familiar.

Who the exam is written for

Microsoft lists six roles for this credential, including Developer and DevOps Engineer, and asks for experience with the software development lifecycle, GitHub workflows and controls, and code quality, security and review practices. It also expects hands-on experience with coding agents such as GitHub Copilot, MCP servers, and agent customisation: custom instructions, custom agents, tools and Copilot setup steps. In plain terms, it is written for someone who already ships software through GitHub and has started handing parts of that work to agents.

NoteNo programming language is named. The exam tests how agents are run, scoped and supervised in a delivery pipeline, not fluency in a particular language.

How to prepare efficiently

Study in weight order, and practise rather than read: most of the blueprint describes configuration you can do in a test repository in an afternoon. Microsoft points to three Learn modules (Foundations of Agentic AI in GitHub; Designing Agent Architecture and SDLC Integration; Tooling, MCP, and Agent Execution Environments) and to GitHub's own documentation for each domain. Then measure yourself against the blueprint to see which domains are actually weak.

NoteExamifyr's free GH-600 readiness test samples all six domains in proportion to their published weights and gives a per-domain breakdown, so you know where to spend time before booking.

Exam tipRead every scenario for who is allowed to do what. Across tool permissions, MCP allow lists, repository and branch scoping, and the guardrails domain, the recurring right answer is the narrowest permission that still lets the agent finish the task, with a human approval only where an action is irreversible or compliance-sensitive.