List your company
ResearchFormerly WizWand

Problem-Solving on MMLU (test)

81.97MMLU Problem Solving AccuracyMANGO
45.070854.650464.2373.8096Aug 18, 2023Jan 31, 2024Jul 16, 2024Dec 29, 2024Jun 14, 2025Nov 27, 2025May 13, 2026
Updated 5mo ago

Evaluation results

MethodLinks
MANGOBase LLM=GPT-4o-mini2026.0581.97
MANGO (Skip-2)Base LLM=GPT-4o-mini2026.0580.73
AFlowBase LLM=GPT-4o-mini2026.0580.52
Plan-and-ExecuteBase LLM=GPT-4o-mini2026.0580.33
DyLANBase LLM=GPT-4o-mini2026.0580.32
MaASBase LLM=GPT-4o-mini2026.0580.3
GPTSwarmBase LLM=GPT-4o-mini2026.0578.34
AgentSquareBase LLM=GPT-4o-mini2026.0578.01
ADASBase LLM=GPT-4o-mini2026.0569.6
MultiPersonaBase LLM=GPT-4o-mini2026.0565.94
SC (CoT×5)Base LLM=GPT-4o-mini2026.0565.9
LLM-DebateBase LLM=GPT-4o-mini2026.0565.64
CoTBase LLM=GPT-4o-mini2026.0565.43
Self-RefineBase LLM=GPT-4o-mini2026.0563.5
IO (direct LLM invocation)Base LLM=GPT-4o-mini2026.0563.44
VICUNA-7BModel size=7B2023.0847.18
STARLING (BLUE)Data strategy=BLUE2023.0846.69
VICUNA-FT-7BModel size=7B, Training=Fine-tuned2023.0846.53
STARLING (BLUE-RED)Data strategy=BLUE-RED2023.0846.49