List your company

Group-level distractor generation on MMLU Elementary Math

34.95RecallDeepseek-V3
17.613222.114126.61531.1159Aug 15, 2025

Evaluation results

MethodLinks
Deepseek-V3Framework=MCTS-guided reasoning reconstruction, Backbone=Deepseek-V32025.0834.95
Claude-4-SonnetFramework=MCTS-guided reasoning reconstruction, Backbone=Claude-4-Sonnet2025.0832.44
LLaMA-3-70BFramework=MCTS-guided reasoning reconstruction, Backbone=LLaMA-3-70B2025.0830.29
GPT-4oFramework=MCTS-guided reasoning reconstruction, Backbone=GPT-4o2025.0830.29
GPT-3.5-turboFramework=MCTS-guided reasoning reconstruction, Backbone=GPT-3.5-turbo2025.0827.96
DiVERT2025.0825.09
O&A2025.0822.22
D-GEN2025.0818.28