Speculative Decoding
Benchmarks
Dataset NameSOTA methodMetricTrendResultsLast Updated
9.43Average Generation Length (τ)
109
Jun 4, 2026
6.06Tau (τ)
81
Jun 4, 2026
7.16Speedup Factor
66
Jun 2, 2026
195.6MT Score
57
May 28, 2026
6.75Speedup
52
Jun 4, 2026
7.99Speedup
52
Jun 4, 2026
6.99Speedup Factor
52
Jun 4, 2026
91.17Throughput
48
May 22, 2026
69.15Throughput (T)
48
May 22, 2026
900.7AVG SR
47
May 27, 2026
6.73Speedup
32
Jun 26, 2026
8.06MAT
30
Apr 24, 2026
1MAT
30
Apr 24, 2026
6.47Match Rate (MAT)
30
Apr 24, 2026
7.1MAT
30
Apr 24, 2026
7.18MAT
30
Apr 24, 2026
7.69Match Rate
30
Apr 24, 2026
7.69MAT
30
Apr 24, 2026
4.83Mean Acceptance Length (tau)
26
Jun 2, 2026
7.08Speedup
26
Jun 2, 2026
128.51Throughput (tokens/s)
22
May 28, 2026
132.69Throughput (tokens/s)
22
May 28, 2026
138.72Throughput (tokens/s)
22
May 28, 2026
2.27GQA Speedup
22
Mar 17, 2026
6.55Mean Expected Tokens per Speculation Step
20
May 5, 2026