Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Long-context evaluation on RULER 128k (matched-train holdout)
Loading...
36.1
Accuracy
Still
16.34
21.47
26.6
31.73
Jun 5, 2026
Accuracy
Updated 1mo ago
Evaluation Results
Method
Method
Links
Accuracy
Still
K (Cache budget)=2048,...
2026.06
36.1
Still
K (Cache budget)=1024,...
2026.06
32.9
Still
K (Cache budget)=164,...
2026.06
31.3
Still
K (Cache budget)=328,...
2026.06
30.2
Still
K (Cache budget)=512,...
2026.06
27.4
KV-Distill
K (Cache budget)=1024,...
2026.06
20.3
KV-Distill
K (Cache budget)=2048,...
2026.06
19.4
KV-Distill
K (Cache budget)=512,...
2026.06
19.4
Still
K (Cache budget)=82, N...
2026.06
19.3
KV-Distill
K (Cache budget)=328,...
2026.06
17.9
KV-Distill
K (Cache budget)=82, N...
2026.06
17.6
KV-Distill
K (Cache budget)=164,...
2026.06
17.1
Feedback
Search any
task
Search any
task