UnbiasedStrata · 12 GB GPU
RX 6700 XT. Native C++ benchmark, prior-suite tasks and v3 Quick Screen. Includes failed admission and retrieval outcomes.
Read report →SONNY’S LABS
Local AI.
Measured on real hardware.
Benchmark reports on model quality, coding, inference speed and the limits of consumer GPUs. Every report identifies its tested configuration, evidence and caveats.
PolyStrata · v3 Quick Screen
Six model configurations. AutomationBench workflows, LiveCodeBench and inference speed.
Read report →PolyStrata · 12 GB GPU
RX 6700 XT. Six model families, practical tasks, HumanEval+ and long-context retrieval.
Read report →Strata · 12 GB GPU
Original Strata v0.1.45 study on the RX 6700 XT. Historical comparison for the later forks.
Read report →Strata v0.1.43 · RTX 5070 Ti
Strata on a 16 GB GPU, with reproducible report data and model/quant configuration details.
Read report →Qwen3.8 · v2.1 comparison
Applied quality and coding evaluation under the earlier Sonny’s Labs v2.1 framework.
Read report →All six indexed reports are hosted in this archive. Benchmark protocols and configurations differ; each report explains the comparisons its evidence supports.