Benchmark local AI hardware. Honestly.

Name: AIHWBench local AI hardware benchmark results
Creator: AIHWBench
License: https://www.apache.org/licenses/LICENSE-2.0

AIHWBench is a vendor-neutral, open platform for measuring how runtimes like Ollama, llama.cpp, ONNX Runtime and OpenVINO perform on real consumer hardware. Every published number comes from a real, reproducible run — missing metrics are shown as “—”, never guessed.

View leaderboard Quick start Methodology

Published results

Hardware configurations

Runtimes covered

Models benchmarked

Measure your machine in one command

Install the CLI, pick a runtime you already have, and run a versioned workload with a fixed seed:

pip install aihwbench && aihwbench doctor

aihwbench benchmark --runtime ollama --model qwen2.5:0.5b-instruct-q4_K_M

Results are saved as JSON with full environment metadata, ready to validate and submit.

Reproducible by design

Fixed seeds, recorded prompts, model checksums, runtime versions and environment snapshots — plus portable .aihwbench bundles with SHA-256 integrity.

Vendor neutral

No vendor pays for placement, no hardware is favored. The leaderboard ranks only what was actually measured.

Privacy first

Local-first: results contain hardware class identifiers only — no serial numbers, paths or personal data. Every submission is privacy-scanned.