List your company

Multi-objective Retrieval on NeedleBench M-RT (32k context stress slice)

100Slot-set ExactnessFullKV
39.36855.10970.8586.591May 7, 2026

Evaluation results

MethodLinks
FullKVBackbone=Qwen3-8B [39], Context=32k2026.0510037.5
FullKVBackbone=Llama-3.1-8B [43], Context=32k2026.0510075
FullKVContract Type=reference2026.0595.837.5
FullKVBackbone=Average, Context=32k2026.0595.837.5
FullKVBackbone=Mistral-7B [44], Context=32k2026.0587.50
MII-QDBackbone=Qwen3-8B [39], Context=32k2026.057550
MII-QDBackbone=Llama-3.1-8B [43], Context=32k2026.057562.5
MII-QDContract Type=question-routed2026.0572.254.2
MII-QDBackbone=Average, Context=32k2026.0572.254.2
MII-pooledBackbone=Qwen3-8B [39], Context=32k2026.0566.737.5
SnapKVBackbone=Llama-3.1-8B [43], Context=32k2026.0566.750
MII-QDBackbone=Mistral-7B [44], Context=32k2026.0566.750
SnapKVBackbone=Qwen3-8B [39], Context=32k2026.0562.525
SnapKVContract Type=fixed contract2026.0561.125
SnapKVBackbone=Average, Context=32k2026.0561.125
SnapKVBackbone=Mistral-7B [44], Context=32k2026.0554.20
MII-pooledContract Type=pooled proxy2026.0552.825
MII-pooledBackbone=Average, Context=32k2026.0552.825
MII-pooledBackbone=Llama-3.1-8B [43], Context=32k2026.055037.5
MII-pooledBackbone=Mistral-7B [44], Context=32k2026.0541.70