Speculative Decoding
Benchmarks
Dataset NameSOTA methodMetricTrendResultsLast Updated
32Max Accepted Token Length
8
Feb 26, 2026
62Spec. Accept %
7
Feb 26, 2026
148.2Throughput (tok/s)
6
Jul 1, 2026
6.99Accepted Length L
6
Jul 1, 2026
7.35Accepted Length L
6
Jul 1, 2026
60MTP Accuracy
6
Jun 11, 2026
2.38Speedup
6
May 25, 2026
3.32Speedup
6
May 25, 2026
167.6Rollout Time (s)
6
May 12, 2026
0.735Speculative Accept Rate
6
Feb 26, 2026
133Throughput (tokens/s)
5
Jun 2, 2026
104.5Throughput (tokens/s)
5
Jun 2, 2026
3.65Overall Mean Score
5
May 27, 2026
392.7Throughput (tokens/s)
5
May 27, 2026
5.12Conversation Score
5
Feb 27, 2026
2.13Block Efficiency
5
Feb 26, 2026
2.32Block Efficiency
5
Feb 26, 2026
2.32Block Efficiency
5
Feb 26, 2026
2.98Speedup
5
Mar 4, 2026
3.68MT Score
4
May 7, 2026
3.74MT Score
4
May 7, 2026
3.52MT Score
4
May 7, 2026
2.3Avg. Acceptance Length (τ)
4
May 7, 2026
2.73MT Score
4
May 7, 2026
3.19MT Score
4
May 7, 2026