training

696.87 category score

Model	Precision	Sustained
MobileNetV3	FP32	275.35
MobileNetV3	BF16	116.95
Flan-T5 Small	FP32	1482.85
Flan-T5 Small	BF16	862.82
DistilGPT2	FP32	1571.28
DistilGPT2	BF16	911.36
DistilBERT	FP32	1583.23
DistilBERT	BF16	1524.76
GraphGPS (Peptides)	FP32	426.90
GraphGPS (Peptides)	BF16	313.78

inference

1285.11 category score

Model	Precision	Sustained
MobileNetV3	FP16	733.83
MobileNetV3	INT8	515.38
Flan-T5 Small	FP16	3458.97
Flan-T5 Small	INT8	1070.69
Flan-T5 Small	FP8	1719.40
DistilGPT2	FP16	3318.55
DistilGPT2	INT8	1084.29
DistilGPT2	FP8	1689.36
DistilBERT	FP16	2816.12
DistilBERT	INT8	946.39
DistilBERT	FP8	1524.79
GraphGPS (Peptides)	FP16	1139.44
GraphGPS (Peptides)	INT8	536.07
GraphGPS (Peptides)	FP8	582.67

compute

1023.25 category score

Model	Precision	Sustained
Dense MatMul	FP32	10955.80
Dense MatMul	FP16	12388.20
Dense MatMul	BF16	12278.69
Dense MatMul	FP64	4320.21
Sparse MatMul	FP32	132.30
Sparse MatMul	FP64	77.41

Provisional results — not the final, official release · GT = LynxBenchAI Global Score. Every precision result is sustained throughput under a declared, bounded optimisation budget — no collapsed single number below the category level.

← Back to the leaderboard