training

60.47 category score

Model	Precision	Sustained
MobileNetV3	FP32	80.92
MobileNetV3	BF16	54.38
Flan-T5 Small	FP32	85.35
Flan-T5 Small	BF16	48.66
DistilGPT2	FP32	72.62
DistilGPT2	BF16	44.78
DistilBERT	FP32	91.89
DistilBERT	BF16	76.34
GraphGPS (Peptides)	FP32	34.04
GraphGPS (Peptides)	BF16	38.21

inference

99.69 category score

Model	Precision	Sustained
MobileNetV3	FP16	258.84
MobileNetV3	INT8	60.95
Flan-T5 Small	FP16	190.70
Flan-T5 Small	INT8	96.98
Flan-T5 Small	FP8	101.92
DistilGPT2	FP16	137.45
DistilGPT2	INT8	77.87
DistilGPT2	FP8	81.55
DistilBERT	FP16	147.25
DistilBERT	INT8	90.94
DistilBERT	FP8	98.62
GraphGPS (Peptides)	FP16	64.47
GraphGPS (Peptides)	INT8	31.39
GraphGPS (Peptides)	FP8	32.19

compute

59.72 category score

Model	Precision	Sustained
Dense MatMul	FP32	448.18
Dense MatMul	FP16	456.50
Dense MatMul	BF16	486.61
Dense MatMul	FP64	13.86
Sparse MatMul	FP32	11.07
Sparse MatMul	FP64	9.23

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