List your company

Question Answering on NQ ID (Accuracy, ECE, Brier, AUROC)

69.3AccuracyCAR (MSCR)
20.4233.1145.858.49Jan 12, 2026

Evaluation results

MethodLinks
CAR (MSCR)Model=Qwen2.5-7B2026.0169.30.2380.2550.637
CAR (Weighted Brier, λ=1/3)Model=Qwen2.5-7B2026.0167.70.2810.2930.629
MASHModel=Qwen2.5-7B2026.01670.3090.3150.543
CAR (Weighted Brier, λ=1)Model=Qwen2.5-7B2026.0165.20.2210.2420.693
Vanilla Search-R1Model=Qwen2.5-7B2026.0161.10.3630.3670.563
Temperature ScalingModel=Qwen2.5-7B2026.0161.10.3110.330.563
Vanilla Search-R1Model=Qwen3-4B2026.0145.70.4520.4380.634
Temperature ScalingModel=Qwen3-4B2026.0145.70.380.3770.634
CAR (MSCR)Model=Qwen2.5-3B2026.0145.50.3030.3030.699
CAR (MSCR)Model=Qwen3-4B2026.0145.50.2720.2720.724
CAR (Weighted Brier, λ=1/3)Model=Qwen2.5-3B2026.0144.50.3070.3130.677
CAR (Weighted Brier, λ=1/3)Model=Qwen3-4B2026.0144.20.2740.2740.718
MASHModel=Qwen2.5-3B2026.0143.40.4790.470.612
Vanilla Search-R1Model=Qwen2.5-3B2026.0143.10.5280.5190.6
Temperature ScalingModel=Qwen2.5-3B2026.0143.10.50.4890.6
MASHModel=Qwen3-4B2026.0136.30.5430.5140.645
CAR (Weighted Brier, λ=1)Model=Qwen3-4B2026.0128.50.1150.120.918
CAR (Weighted Brier, λ=1)Model=Qwen2.5-3B2026.0122.30.0910.0910.941