training

511.71 category score

Model	Precision	Sustained
MobileNetV3	FP32	565.29
MobileNetV3	BF16	313.42
Flan-T5 Small	FP32	765.71
Flan-T5 Small	BF16	456.20
DistilGPT2	FP32	728.59
DistilGPT2	BF16	402.79
DistilBERT	FP32	930.51
DistilBERT	BF16	736.92
GraphGPS (Peptides)	FP32	273.57
GraphGPS (Peptides)	BF16	280.73

inference

969.64 category score

Model	Precision	Sustained
MobileNetV3	FP16	2622.46
MobileNetV3	INT8	663.92
Flan-T5 Small	FP16	2079.21
Flan-T5 Small	INT8	1127.07
Flan-T5 Small	FP8	1153.78
DistilGPT2	FP16	1176.82
DistilGPT2	INT8	706.70
DistilGPT2	FP8	709.12
DistilBERT	FP16	1485.98
DistilBERT	INT8	922.77
DistilBERT	FP8	980.32
GraphGPS (Peptides)	FP16	537.23
GraphGPS (Peptides)	INT8	289.02
GraphGPS (Peptides)	FP8	276.63

compute

279.54 category score

Model	Precision	Sustained
Dense MatMul	FP32	3578.56
Dense MatMul	FP16	4215.79
Dense MatMul	BF16	4275.35
Dense MatMul	FP64	101.39
Sparse MatMul	FP32	30.17
Sparse MatMul	FP64	21.19

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