List your company

Decode Latency Benchmark

6.2Decode Latency (ms/tok)Dense Softmax 1.3B
3.8419.7735.751.63May 10, 2026

Evaluation results

MethodLinks
Dense Softmax 1.3BContext=256, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.056.2—
Dense Softmax 1.3BContext=1,024, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.056.912
Dense Softmax 1.3BContext=4,096, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.059.655
Dense Softmax 1.3BContext=8,192, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.0512.9108
Dense Softmax 1.3BContext=16,384, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.0519.6216
Dense Softmax 1.3BContext=24,576, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.0526.5328
MoL Hybrid 2.08BContext=8,192, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.0562.5—
MoL Hybrid 2.08BContext=256, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.0563.2—
MoL Hybrid 2.08BContext=4,096, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.0563.3—
MoL Hybrid 2.08BContext=1,024, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.0564.2—
MoL Hybrid 2.08BContext=24,576, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.0565—
MoL Hybrid 2.08BContext=16,384, Batch Size=1, Hardware=RTX 3090, KV Cache Status=Enabled2026.0565.2—