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

GH-600 vs GH-300: GitHub Agentic AI Developer or GitHub Copilot?

Both are intermediate GitHub exams with a 700 pass mark. One certifies using Copilot well; the other certifies running and governing agents in a delivery pipeline.

The short answer

GH-300 (GitHub Copilot) certifies that you use Copilot well: its features, prompting, responsible use and privacy settings. GH-600 (Developing in Agentic AI Systems) certifies that you can run agents as part of software delivery: scoping and permissions, MCP servers, CI-invoked agents, evaluation, multi-agent coordination and guardrails. GH-300 is about you working with an assistant. GH-600 is about agents working in your pipeline, and you supervising them.

Side by side

Every figure below is from the two certification pages and study guides on Microsoft Learn.

                  GH-300 (Copilot)          GH-600 (Agentic AI Dev)
Level             Intermediate              Intermediate
Duration          100 minutes               120 minutes
Passing           700, scaled               700, scaled
Languages         EN, ES, PT-BR, KO, JA     English only
Practice test     Free, on Microsoft Learn  None yet
Valid for         2 years                   2 years
Price             set per country           set per country

What each blueprint weights

GH-300's weight sits in using Copilot's features (25-30%), with responsible use at 15-20% and four more areas at 10-15% each: data and architecture, prompt engineering, developer productivity, and privacy and content exclusions. Microsoft's current GH-300 study guide lists the features domain twice under slightly different names; it is one domain. GH-600 has six domains, every one about agents, led by tool use and environment interaction at 20-25%.

GH-300                                       GH-600
Use GitHub Copilot features ....... 25-30%   Tool use & environment ........ 20-25%
Use GitHub Copilot responsibly .... 15-20%   Agent architecture & SDLC ..... 15-20%
Copilot data and architecture ..... 10-15%   Evaluation, errors & tuning ... 15-20%
Prompt engineering & context ...... 10-15%   Multi-agent coordination ...... 15-20%
Developer productivity ............ 10-15%   Memory, state & execution ..... 10-15%
Privacy, exclusions & safeguards .. 10-15%   Guardrails & accountability ... 10-15%

Where they overlap

Both touch agents and MCP. GH-300 includes using agent mode, MCP, agent sessions and delegating to sub-agents inside your own workflow. GH-600 goes further: configuring a GitHub remote MCP server, MCP registries and allow lists, scoping agents to repositories and branches, invoking them from CI, and building retries, rollbacks and escalation paths around them. If the GH-300 material on agents felt like the most interesting part, GH-600 is that part expanded into a whole exam.

Which one to take

Take GH-300 if you want to show you use Copilot effectively and responsibly as an individual developer, or if you prefer to sit the exam in Spanish, Portuguese, Korean or Japanese. Take GH-600 if agents already open pull requests or run in CI on your team, or if you are the person deciding what they are allowed to do. Microsoft does not require GH-300 before GH-600, but its GH-600 audience profile expects hands-on experience with coding agents including Copilot, which is roughly what GH-300 certifies.

Exam tipIf you are unsure, read the GH-600 tool-use and guardrails bullets. If they describe decisions you have had to make about real repositories, take GH-600. If they describe things nobody on your team has done yet, take GH-300 first.
← All GitHub GH-600 guides