Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Reasoning step reduction on In-Distribution 5K corpus (test)
Loading...
47.5
Savings Rate
TTT no-QK
28.78
33.64
38.5
43.36
Apr 1, 2026
Savings Rate
Error Rate
Updated 16d ago
Evaluation Results
Method
Method
Links
Savings Rate
Error Rate
TTT no-QK
Backbone=Qwen2.5-32B,...
2026.04
47.5
11
TTT no-QK
Backbone=Llama-3.3-70B...
2026.04
42.4
9
TTT QK
Backbone=Qwen2.5-32B,...
2026.04
41.4
10.3
TTT no-QK
Backbone=QwQ-32B, Supe...
2026.04
39.4
8.1
Static Probe
Backbone=Qwen2.5-32B,...
2026.04
38
10.5
TTT QK
Backbone=Llama-3.3-70B...
2026.04
37.8
8.1
TTT QK
Backbone=QwQ-32B, Supe...
2026.04
37.6
7.6
Static Probe
Backbone=Llama-3.3-70B...
2026.04
35.4
10.4
Static Probe
Backbone=QwQ-32B, Supe...
2026.04
29.5
9.4
Feedback
Search any
task
Search any
task