Loading...
Loading...
Standardized benchmarks for OpenClaw forks running on real hardware constraints. Each test runs in a Docker container matching the device's CPU and memory profile.
| # | Device | Fork | Score |
|---|---|---|---|
| 1 | Mac Mini M1 (used)Mini PC | PythonNanobot | 91 |
| 2 | Beelink SER5 Max |
A repeatable, containerized pipeline that tests every fork under real hardware constraints.
Fork is cloned into a Docker container with CPU and RAM limits matching the target device.
Dependencies installed via the native toolchain — Go, Rust, Python, TypeScript, or C.
Entry point detected, cold start timed, peak memory tracked via cgroup, disk usage measured.
Results combined into a 0–100 composite score weighted across four dimensions.
| RustIronClaw |
| 80 |
| 3 | Mac Mini M1 (used)Mini PC | TypeScriptNanoClaw | 80 |
| 4 | Beelink Mini S12 ProMini PC | TypeScriptNanoClaw | 80 |
| 5 | Mac Mini M1 (used)Mini PC | TypeScriptOpenClaw | 78 |
| 6 | Geekom Mini IT12Mini PC | TypeScriptNanoClaw | 70 |
| 7 | ASUS PN53Mini PC | TypeScriptNanoClaw | 70 |
| 8 | GMKtec NucBox K8Mini PC | TypeScriptNanoClaw | 65 |
| 9 | Lenovo ThinkCentre M75q TinyMini PC | TypeScriptNanoClaw | 65 |
| 10 | LattePanda SigmaMini PC | TypeScriptNanoClaw | 60 |
| 11 | Beelink Mini S12 ProMini PC | TypeScriptOpenClaw | 50 |
| 12 | ASUS PN53Mini PC | TypeScriptOpenClaw | 50 |
| 13 | Intel NUC 13 ProMini PC | TypeScriptOpenClaw | 50 |
| 14 | Geekom Mini IT12Mini PC | TypeScriptOpenClaw | 50 |
| 15 | GMKtec NucBox K8Mini PC | TypeScriptOpenClaw | 50 |
| 16 | Lenovo ThinkCentre M75q TinyMini PC | TypeScriptOpenClaw | 50 |
| 17 | LattePanda SigmaMini PC | TypeScriptOpenClaw | 50 |
| 18 | Beelink Mini S12 ProMini PC | RustIronClaw | 50 |
| 19 | Mac Mini M1 (used)Mini PC | RustIronClaw | 50 |
Messaging, browser, code exec, memory, files, search, MCP, tool use
Cold start time — clone + install + startup. Under 5s = full marks
Disk footprint after install. Under 20MB = full marks
5 pts for dependency install, 5 pts for successful startup
Detected via static source analysis and runtime module probing. Each passed test contributes 5 pts to the capabilities score.
Run benchmarks locally or contribute improvements