Our new X account is live! Follow @wizwand_team for updates
Home
/
Benchmarks
Long-context Retrieval on Needle-in-a-Haystack (NiH)
Loading...
100
Accuracy (512 tokens)
Dense
22
42.25
62.5
82.75
Dec 8, 2025
Accuracy (512 tokens)
Accuracy (1024 tokens)
Accuracy (1536 tokens)
Accuracy (2048 tokens)
Accuracy (2560 tokens)
Accuracy (3072 tokens)
Updated 4d ago
Evaluation Results
Method
Method
Links
Accuracy (512 tokens)
Accuracy (1024 tokens)
Accuracy (1536 tokens)
Accuracy (2048 tokens)
Accuracy (2560 tokens)
Accuracy (3072 tokens)
Dense
Prune%=0%, Base Model=...
2025.12
100
100
90
95
45
35
Token Filtering
Prune%=20%, Base Model...
2025.12
70
70
65
55
30
10
FLAP
Prune%=20%, Base Model...
2025.12
25
20
10
0
0
0
Feedback
Search any
task
Search any
task