Local LLM Leaderboard
Local model benchmarks, measured in Doha
We benchmark open-weight models running fully on local hardware — no data leaves the machine. Throughput, latency, footprint, and quality, compared side by side.
Coming soon
We're finalising our benchmarking methodology and first set of results. Check back shortly, or get in touch to suggest a model or hardware target.
Suggest a modelThroughput
Tokens generated per second under sustained load.
First-token latency
Time to first token after prompt submission.
Footprint
Parameters, quantization, and on-disk / memory size.
Quality
Task scoring across reasoning, coding, and bilingual EN/AR prompts.