GitHub-Hosted Runners Reference (2025)

This reference covers all GitHub-hosted runner types, including recent additions and deprecations.

Standard Runner Labels

Ubuntu

runs-on: ubuntu-latest      # Ubuntu 24.04 (default)
runs-on: ubuntu-24.04       # Ubuntu 24.04
runs-on: ubuntu-22.04       # Ubuntu 22.04
runs-on: ubuntu-20.04       # Ubuntu 20.04

Windows

runs-on: windows-latest     # Windows Server 2022 (default)
runs-on: windows-2025       # Windows Server 2025 (NEW)
runs-on: windows-2022       # Windows Server 2022
runs-on: windows-2019       # Windows Server 2019

macOS

runs-on: macos-latest       # macOS 15 (as of Aug 2025)
runs-on: macos-15           # macOS 15 Sequoia (Apple Silicon)
runs-on: macos-14           # macOS 14 Sonoma (Apple Silicon)
runs-on: macos-26           # macOS 26 (PREVIEW)

---

macOS Runner Updates and Deprecations

Current Status

Intel-Specific Labels (Long-term Deprecated)

runs-on: macos-15-intel     # Intel x86_64 (NEW but deprecated long-term)
runs-on: macos-14-large     # Intel x86_64
runs-on: macos-15-large     # Intel x86_64

Important: Apple Silicon (ARM64) will be required after Fall 2027. Plan migration now.

Migration Example

jobs:
  build:
    # BAD - macos-13 retired Nov 14, 2025 (WILL FAIL)
    # runs-on: macos-13

    # GOOD - Use macos-15 or macos-latest
    runs-on: macos-15

    steps:
      - uses: actions/checkout@v6
      - run: ./build.sh

---

ARM64 Runners

Availability

• Generally available as of August 2025
• Free for public repositories
• Private repositories require GitHub Enterprise Cloud plan

Labels

runs-on: ubuntu-latest-arm64    # Free for public repos
runs-on: ubuntu-24.04-arm64
runs-on: windows-latest-arm64   # ARM Windows

Example

jobs:
  build:
    runs-on: ubuntu-latest-arm64  # Free for public repos
    steps:
      - uses: actions/checkout@v6
      - name: Build on ARM64
        run: |
          uname -m  # Should output: aarch64
          ./build.sh

Specifications

• 4 vCPU ARM64 processors
• Available for Linux and Windows
• Native ARM execution (no virtualization needed)
• Ideal for multi-architecture builds

Common Validation Issues

• Using ARM64 runners in private repos without Enterprise Cloud
• Assuming all community actions work on ARM64
• Not testing ARM-specific compilation issues

---

GPU Runners

Availability

• Generally available for Windows and Linux
• Requires Team or Enterprise Cloud plan

Labels

runs-on: gpu-t4-4-core      # NVIDIA Tesla T4

Specifications

• GPU: NVIDIA Tesla T4 with 16GB VRAM
• CPU: 4 vCPUs
• RAM: 28GB
• Pricing: $0.07/minute

Example

jobs:
  ml-training:
    runs-on: gpu-t4-4-core
    steps:
      - uses: actions/checkout@v6

      - name: Setup Python
        uses: actions/setup-python@v5
        with:
          python-version: '3.11'

      - name: Install ML dependencies
        run: |
          pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
          pip install -r requirements.txt

      - name: Train model
        run: python train.py --use-gpu

      - name: Run inference
        run: python inference.py

Use Cases

• ML model training
• GPU-accelerated testing
• CUDA development
• Image/video processing

Common Validation Issues

• Missing CUDA setup
• Incorrect GPU driver versions
• Not utilizing GPU in workloads (CPU fallback)
• Missing GPU-specific dependencies

---

M2 Pro macOS Runners (Larger Runners)

Availability

• Generally available with M2 Pro powered runners

Labels

runs-on: macos-latest-xlarge  # macOS 15, M2 Pro
runs-on: macos-15-xlarge      # macOS 15, M2 Pro
runs-on: macos-14-xlarge      # macOS 14, M2 Pro

Specifications

• CPU: 5-core (vs 3-core standard)
• GPU: 8-core with hardware acceleration (enabled by default)
• RAM: 14GB
• Storage: 14GB
• Performance: Up to 15% faster than M1 runners
• Pricing: $0.16/minute

Example

jobs:
  ios-build:
    runs-on: macos-15-xlarge  # M2 Pro with GPU acceleration
    steps:
      - uses: actions/checkout@v6

      - name: Setup Xcode
        uses: maxim-lobanov/setup-xcode@v1
        with:
          xcode-version: latest-stable

      - name: Build iOS app
        run: |
          xcodebuild -workspace App.xcworkspace \\
            -scheme Production \\
            -configuration Release \\
            -archivePath build/App.xcarchive \\
            archive

      - name: Run GPU-accelerated tests
        run: |
          # GPU acceleration automatically available
          xcodebuild test -scheme AppTests

Benefits

• GPU hardware acceleration for Metal-based workloads
• Improved build times for iOS/macOS apps
• Better performance for Xcode builds
• Native Apple Silicon performance

---

Runner Selection Best Practices

Decision Criteria

1. Architecture compatibility: ARM64 vs Intel x86_64
2. Cost optimization: Standard vs larger runners
3. GPU requirements: ML/AI workloads need GPU runners
4. Operating system: Latest versions recommended
5. Deprecation timelines: Avoid retired runners
6. Public vs private repos: ARM64 free only for public repos

Validation Checklist

# Check these in your workflows:
- [ ] Using latest runner versions (macos-15, windows-2025, ubuntu-latest)
- [ ] Not using deprecated runners (macos-13)
- [ ] Architecture-appropriate runners (ARM64 vs Intel)
- [ ] GPU runners for ML workloads
- [ ] Cost-effective runner selection
- [ ] ARM64 compatibility tested (if using ARM64 runners)

Cost Comparison

---

Multi-Architecture Builds

Example: Building for Multiple Architectures

jobs:
  build:
    strategy:
      matrix:
        include:
          - runner: ubuntu-latest
            arch: x64
          - runner: ubuntu-latest-arm64
            arch: arm64

    runs-on: ${{ matrix.runner }}
    steps:
      - uses: actions/checkout@v6

      - name: Build
        run: |
          echo "Building for ${{ matrix.arch }}"
          ./build.sh --arch ${{ matrix.arch }}

      - name: Upload artifact
        uses: actions/upload-artifact@v4
        with:
          name: build-${{ matrix.arch }}
          path: dist/

---

Self-Hosted Runner Configuration

When using self-hosted runners, configure actionlint to recognize custom labels:

# .github/actionlint.yaml
self-hosted-runner:
  labels:
    - my-custom-runner
    - gpu-runner
    - arm-runner
    - on-premises

This prevents actionlint from reporting unknown runner label errors.