Benchmarks
Model benchmarks.
Local models on a 2× NVIDIA DGX Spark cluster. StoryBench asks: can it still write, and which world do the stories inhabit? DharmaBench asks: what does it actually believe?
Head to head · 2 quants, 3 belts
Qwen35-A3B Quant Duel
NVIDIA NVFP4 vs unsloth NVFP4-Fast — same base model, same hardware, same judges, one page.
Enter the arena →
Benchmark queue
What runs next on the cluster. Every model gets the full treatment: serve on both nodes, StoryBench, DharmaBench, publish here.