training

700.38 category score

Model	Precision	Sustained
MobileNetV3	FP32	564.13
MobileNetV3	BF16	257.12
Flan-T5 Small	FP32	1095.97
Flan-T5 Small	BF16	622.63
DistilGPT2	FP32	1159.01
DistilGPT2	BF16	651.85
DistilBERT	FP32	1480.99
DistilBERT	BF16	1291.98
GraphGPS (Peptides)	FP32	374.00
GraphGPS (Peptides)	BF16	386.94

inference

1659.36 category score

Model	Precision	Sustained
MobileNetV3	FP16	2832.98
MobileNetV3	INT8	931.43
Flan-T5 Small	FP16	3919.63
Flan-T5 Small	INT8	1389.31
Flan-T5 Small	FP8	1658.94
DistilGPT2	FP16	3797.65
DistilGPT2	INT8	1489.31
DistilGPT2	FP8	1920.23
DistilBERT	FP16	2611.72
DistilBERT	INT8	1090.29
DistilBERT	FP8	1362.71
GraphGPS (Peptides)	FP16	1060.04
GraphGPS (Peptides)	INT8	520.52
GraphGPS (Peptides)	FP8	548.07

compute

471.03 category score

Model	Precision	Sustained
Dense MatMul	FP32	10266.87
Dense MatMul	FP16	12642.72
Dense MatMul	BF16	13150.56
Dense MatMul	FP64	4143.71
Sparse MatMul	FP32	16.10
Sparse MatMul	FP64	28.04

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