List your company

In-context retrieval on DROP

88.6AccuracyBERT-Judge
11.6431.6251.671.58Mar 15, 2026Mar 19, 2026Mar 23, 2026Mar 28, 2026Apr 1, 2026Apr 5, 2026Apr 10, 2026

Evaluation results

MethodLinks
BERT-Judgeclean_name=BERT-as-a-Judge2026.0488.6
Regex2026.0477
LLM-Judgebackbone=Qwen-3 0.6B2026.0469.3
Transformer++Parameter scale=410M2026.0328.2
Hybrid Gated DeltaNet + M2RNN-3Parameter scale=410M2026.0327.2
Hybrid M2RNNParameter scale=410M2026.0326.8
Hybrid Mamba-2Parameter scale=410M2026.0326.4
Hybrid Mamba-2 + M2RNN-3Parameter scale=410M2026.0326.4
Hybrid Mamba-2 + M2RNN-1Parameter scale=410M2026.0326.3
Hybrid Gated DeltaNetParameter scale=410M2026.0326
Hybrid Gated DeltaNet + M2RNN-1Parameter scale=410M2026.0325.6
Gated DeltaNetParameter scale=410M2026.0324.9
M2RNNParameter scale=410M2026.0322.9
Mamba-2Parameter scale=410M2026.0319.5
GRUParameter scale=410M2026.0317.2
RNNParameter scale=410M2026.0314.6