Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Long-context evaluation on RULER 64k (matched-train holdout)
Loading...
39.9
Accuracy
Still
19.412
24.731
30.05
35.369
Jun 5, 2026
Accuracy
Updated 1mo ago
Evaluation Results
Method
Method
Links
Accuracy
Still
K (Cache budget)=2048,...
2026.06
39.9
Still
K (Cache budget)=1024,...
2026.06
36.5
Still
K (Cache budget)=328,...
2026.06
36.5
Still
K (Cache budget)=164,...
2026.06
35.1
Still
K (Cache budget)=512,...
2026.06
31.8
Still
K (Cache budget)=82, N...
2026.06
24.2
KV-Distill
K (Cache budget)=82, N...
2026.06
21.9
KV-Distill
K (Cache budget)=512,...
2026.06
21.1
KV-Distill
K (Cache budget)=2048,...
2026.06
21
KV-Distill
K (Cache budget)=1024,...
2026.06
20.4
KV-Distill
K (Cache budget)=328,...
2026.06
20.4
KV-Distill
K (Cache budget)=164,...
2026.06
20.2
Feedback
Search any
task
Search any
task