Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Long-context evaluation on RULER 32k (matched-train holdout)
Loading...
48.7
Accuracy
Still
17.5
25.6
33.7
41.8
Jun 5, 2026
Accuracy
Updated 1mo ago
Evaluation Results
Method
Method
Links
Accuracy
Still
K (Cache budget)=2048,...
2026.06
48.7
Still
K (Cache budget)=1024,...
2026.06
45.2
Still
K (Cache budget)=328,...
2026.06
39.2
Still
K (Cache budget)=512,...
2026.06
38.4
Still
K (Cache budget)=164,...
2026.06
34.5
Still
K (Cache budget)=82, N...
2026.06
29.1
KV-Distill
K (Cache budget)=2048,...
2026.06
26.5
KV-Distill
K (Cache budget)=1024,...
2026.06
24
KV-Distill
K (Cache budget)=512,...
2026.06
22.1
KV-Distill
K (Cache budget)=328,...
2026.06
21.2
KV-Distill
K (Cache budget)=164,...
2026.06
19
KV-Distill
K (Cache budget)=82, N...
2026.06
18.7
Feedback
Search any
task
Search any
task