Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Mathematical Reasoning on GSM8K (Acc, TNFT, Tokens, Delay)
Loading...
91.85
Accuracy
Base
88.73
89.54
90.35
91.16
May 12, 2026
Accuracy
Time to First Token (TNFT)
Output Token Count
Inference Delay
Updated 21d ago
Evaluation Results
Method
Method
Links
Accuracy
Time to First Token (TNFT)
Output Token Count
Inference Delay
Base
Backbone=Qwen3-4B
2026.05
91.85
117.03
1,340.49
41.94
Vanilla
Backbone=Qwen3-1.7B
2026.05
90.6
93.3
660.64
14.93
Base
Backbone=Qwen3-1.7B
2026.05
90.37
117.03
1,156.1
27.14
Ours
Backbone=Qwen3-1.7B
2026.05
89.51
0
437.1
11.29
Vanilla
Backbone=Qwen3-4B
2026.05
89.36
93.3
649.28
20.17
Ours
Backbone=Qwen3-4B
2026.05
88.85
0
421.47
14.53
Feedback
Search any
task
Search any
task