# Benchmarks (https://luxor.ai/docs/runners/benchmarks)

> For the complete documentation index, see [llms.txt](https://luxor.ai/llms.txt)

Benchmark results and methodology comparing Runners against GitHub-hosted runners across real open-source build workloads.

Our runners run on modern bare-metal hardware, so the same jobs finish faster than on GitHub-hosted runners with the same number of cores. The per-minute rate is lower too, so each run costs less.

## How we benchmarked

* Runners: `tenki-standard-large-8c-16g` (8 vCPU, 16 GB) against GitHub's 8-core larger runner (8 vCPU, 32 GB). Kubernetes ran on 16 vCPU on both sides.
* Workloads: each project's own CI job or standard build at a pinned commit, with identical steps on both sides and only `runs-on` changed.
* Caches: dependency caches are warm on both sides. Build caches (sccache, Docker layers, the Gradle build cache) are off, so every run compiles in full.
* Timing: the whole job as GitHub reports it, including setup and checkout.
* Runs: the suite ran several times on November 11, 2026. Each row is the project's fastest run relative to GitHub among runs where both sides succeeded.

## Results vs GitHub-hosted runners

| Workload                                      | What runs                               | GitHub-hosted | Luxor Agent Cloud | Speedup     |
| --------------------------------------------- | --------------------------------------- | ------------- | ----------------- | ----------- |
| Docker build (`PostHog/posthog`)              | `docker buildx` of the production image | 14m 30s       | 4m 33s            | 3.2x faster |
| Node.js (`nodejs/node`)                       | Build and full test suite               | 1h 50m 41s    | 37m 15s           | 3.0x faster |
| Linux kernel (`torvalds/linux`)               | `make defconfig`, then `make -j8`       | 7m 1s         | 2m 56s            | 2.4x faster |
| Apache Kafka (`apache/kafka`)                 | `./gradlew build -x test`               | 8m 34s        | 3m 48s            | 2.3x faster |
| Spring Boot (`spring-projects/spring-boot`)   | `./gradlew build`                       | 1h 22m 58s    | 47m 1s            | 1.8x faster |
| Kubernetes (`kubernetes/kubernetes`, 16 vCPU) | `make`, then `make test`                | 19m 17s       | 11m 43s           | 1.6x faster |

To have a specific workload reproduced, write to [hello@luxor.ai](mailto:hello@luxor.ai).

## Cost

| Workload     | GitHub-hosted | Luxor Agent Cloud | Difference |
| ------------ | ------------- | ----------------- | ---------- |
| Docker build | $0.330        | $0.073            | 78% less   |
| Node.js      | $2.442        | $0.596            | 76% less   |
| Linux kernel | $0.176        | $0.047            | 73% less   |
| Apache Kafka | $0.198        | $0.061            | 69% less   |
| Spring Boot  | $1.826        | $0.752            | 59% less   |
| Kubernetes   | $0.840        | $0.375            | 55% less   |

GitHub charges $0.022/min for its 8-core runner and $0.042/min for 16-core, and rounds each job up to the next whole minute. Luxor Agent Cloud charges $0.002 per vCPU-minute, billed per second with no startup surcharge. See [the pricing page](/pricing) for the full rate card.