training

8.47 category score

Model	Precision	Sustained
MobileNetV3	FP32	26.84
MobileNetV3	BF16	6.89
Flan-T5 Small	FP32	11.27
Flan-T5 Small	BF16	6.02
DistilGPT2	FP32	9.04
DistilGPT2	BF16	5.00
DistilBERT	FP32	9.95
DistilBERT	BF16	6.24
GraphGPS (Peptides)	FP32	7.45
GraphGPS (Peptides)	BF16	3.06

inference

4.88 category score

Model	Precision	Sustained
MobileNetV3	FP16	41.47
MobileNetV3	INT8	21.89
Flan-T5 Small	FP16	5.81
Flan-T5 Small	INT8	18.58
Flan-T5 Small	FP8	0.00
DistilGPT2	FP16	5.45
DistilGPT2	INT8	8.33
DistilGPT2	FP8	0.00
DistilBERT	FP16	5.12
DistilBERT	INT8	9.21
DistilBERT	FP8	0.00
GraphGPS (Peptides)	FP16	1.15
GraphGPS (Peptides)	INT8	0.33
GraphGPS (Peptides)	FP8	0.00

compute

4.07 category score

Model	Precision	Sustained
Dense MatMul	FP32	43.58
Dense MatMul	FP16	7.66
Dense MatMul	BF16	21.27
Dense MatMul	FP64	8.47
Sparse MatMul	FP32	0.74
Sparse MatMul	FP64	0.90

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