training

332.78 category score

Model	Precision	Sustained
MobileNetV3	FP32	394.16
MobileNetV3	BF16	268.26
Flan-T5 Small	FP32	506.61
Flan-T5 Small	BF16	276.10
DistilGPT2	FP32	469.76
DistilGPT2	BF16	272.36
DistilBERT	FP32	623.27
DistilBERT	BF16	472.39
GraphGPS (Peptides)	FP32	158.56
GraphGPS (Peptides)	BF16	151.20

inference

648.22 category score

Model	Precision	Sustained
MobileNetV3	FP16	1519.03
MobileNetV3	INT8	437.31
Flan-T5 Small	FP16	1333.43
Flan-T5 Small	INT8	650.15
Flan-T5 Small	FP8	778.03
DistilGPT2	FP16	957.02
DistilGPT2	INT8	577.05
DistilGPT2	FP8	617.67
DistilBERT	FP16	1069.74
DistilBERT	INT8	561.18
DistilBERT	FP8	673.12
GraphGPS (Peptides)	FP16	351.19
GraphGPS (Peptides)	INT8	167.13
GraphGPS (Peptides)	FP8	173.88

compute

361.93 category score

Model	Precision	Sustained
Dense MatMul	FP32	2399.40
Dense MatMul	FP16	2954.14
Dense MatMul	BF16	2837.85
Dense MatMul	FP64	94.80
Sparse MatMul	FP32	59.85
Sparse MatMul	FP64	66.62

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