List your company
ResearchFormerly WizWand

General Reasoning on BBH

88.7ScoreDeepSeek V3.2
66.13271.99177.8583.709Feb 11, 2026Feb 14, 2026Feb 18, 2026Feb 21, 2026Feb 25, 2026Feb 28, 2026Mar 4, 2026
Updated 7mo ago

Evaluation results

MethodLinks
DeepSeek V3.2Model Variant=Exp Base, # Shots=3-shot, # Activated Params=37B, # Total Params=671B2026.0288.7———
Kimi-K2Model Variant=Base, # Shots=3-shot, # Activated Params=32B, # Total Params=1043B2026.0288.7———
MiMo-V2 FlashModel Variant=Base, # Shots=3-shot, # Activated Params=15B, # Total Params=309B2026.0288.5———
Step 3.5 FlashModel Variant=Base, # Shots=3-shot, # Activated Params=11B, # Total Params=196B2026.0288.2———
DeepSeek V3.1Model Variant=Base, # Shots=3-shot, # Activated Params=37B, # Total Params=671B2026.0288.2———
GLM-4.5Model Variant=Base, # Shots=3-shot, # Activated Params=32B, # Total Params=355B2026.0286.2———
System PromptBase Model=Llama-3.1-8B2026.0369———
Llama-3.1-8B-SFTBase Model=Llama-3.1-8B2026.0368———
Self-CDBase Model=Llama-3.1-8B2026.0368———
VCLBase Model=Llama-3.1-8B2026.0368———
DROBase Model=Llama-3.1-8B2026.0367———
Vector AblationBase Model=Llama-3.1-8B2026.0367———
Dare Task ArithmeticModels=Fuse2026.02———53.66
Dare Ties MergingModels=Fuse2026.02———55.59
MathCoder2-Mistral-7BModels=base12026.02———46.66
Mistral-7B-Instruct-v0.2Models=base02026.02———47.38
P(True)Confidence Method=P(True), Budget=52025.02—48100—
P(True)Confidence Method=P(True), Budget=102025.02—4790—
Response ProbabilityConfidence Method=Response Probability, Budget=52025.02—3270—
Response ProbabilityConfidence Method=Response Probability, Budget=102025.02—4580—
SCEModels=Fuse2026.02———46.67
SCF-RKLModels=Fuse2026.02———54.47
Task ArithmeticModels=Fuse2026.02———53.66
Ties MergingModels=Fuse2026.02———46.6
Verbal 1-100Confidence Method=Verbal 1-100, Budget=52025.02—2250—
Verbal 1-100Confidence Method=Verbal 1-100, Budget=102025.02—4140—
Verbal BinaryConfidence Method=Verbal Binary, Budget=52025.02—1720—
Verbal BinaryConfidence Method=Verbal Binary, Budget=102025.02—1010—