List your company
ResearchFormerly WizWand

Mathematical Reasoning on MinervaMath (Accuracy)

82.8AccuracyAgent 1 (Qwen3-30B-A3B-Instruct)
2.223.12544.0564.975Sep 28, 2025Nov 7, 2025Dec 17, 2025Jan 27, 2026Mar 8, 2026Apr 17, 2026May 28, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Agent 1 (Qwen3-30B-A3B-Instruct)Controller=Qwen2.5-7B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Qwen2.5-7B-Instruct, Agent 3=Qwen2.5-1.5B-Instruct2026.0582.8
Agent 1 (Qwen3-30B-A3B-Instruct)Controller=Qwen3-4B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Ministral-3-8B-Instruct, Agent 3=Llama-3-2-1B-Instruct2026.0582.8
Iterative Critique-and-Routing ControllerController=Qwen3-4B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Ministral-3-8B-Instruct, Agent 3=Llama-3-2-1B-Instruct2026.0571.8
Iterative Critique-and-Routing ControllerController=Qwen2.5-7B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Qwen2.5-7B-Instruct, Agent 3=Qwen2.5-1.5B-Instruct2026.0567.8
Agent 2 (Ministral-3-8B-Instruct)Controller=Qwen3-4B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Ministral-3-8B-Instruct, Agent 3=Llama-3-2-1B-Instruct2026.0559.9
Random RouterController=Qwen2.5-7B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Qwen2.5-7B-Instruct, Agent 3=Qwen2.5-1.5B-Instruct2026.0553.7
Random RouterController=Qwen3-4B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Ministral-3-8B-Instruct, Agent 3=Llama-3-2-1B-Instruct2026.0552.8
Agent 2 (Qwen2.5-7B-Instruct)Controller=Qwen2.5-7B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Qwen2.5-7B-Instruct, Agent 3=Qwen2.5-1.5B-Instruct2026.0552.4
Controller V1Controller=Qwen3-4B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Ministral-3-8B-Instruct, Agent 3=Llama-3-2-1B-Instruct2026.0546
GAPGModel=LLaMA-3.2-3B2025.0943.4
Router-R1Controller=Qwen2.5-7B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Qwen2.5-7B-Instruct, Agent 3=Qwen2.5-1.5B-Instruct2026.0543.2
Controller V2Controller=Qwen2.5-7B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Qwen2.5-7B-Instruct, Agent 3=Qwen2.5-1.5B-Instruct2026.0542.3
RouterDCController=Qwen2.5-7B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Qwen2.5-7B-Instruct, Agent 3=Qwen2.5-1.5B-Instruct2026.0541.9
Controller V2Controller=Qwen3-4B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Ministral-3-8B-Instruct, Agent 3=Llama-3-2-1B-Instruct2026.0541.9
RoBERTa RouterController=Qwen2.5-7B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Qwen2.5-7B-Instruct, Agent 3=Qwen2.5-1.5B-Instruct2026.0541.7
GAPGModel=Qwen2.5-1.5B2025.0940.8
Router-R1Controller=Qwen3-4B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Ministral-3-8B-Instruct, Agent 3=Llama-3-2-1B-Instruct2026.0540.5
Router-R1Model=LLaMA-3.2-3B2025.0939.7
RouterDCController=Qwen3-4B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Ministral-3-8B-Instruct, Agent 3=Llama-3-2-1B-Instruct2026.0538.6
DAPO + HistaModel Size=14B, RL Algorithm=DAPO2026.0538.5
DAPOModel Size=14B, RL Algorithm=DAPO2026.0538.1
Router-R1Model=Qwen2.5-1.5B2025.0937.8
RoBERTa RouterController=Qwen3-4B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Ministral-3-8B-Instruct, Agent 3=Llama-3-2-1B-Instruct2026.0537.6
CSIPO + HistaModel Size=7B, RL Algorithm=CSIPO2026.0534.2
DAPO + HistaModel Size=7B, RL Algorithm=DAPO2026.0533.1
AutoMixModel=LLaMA-3.2-3B2025.0932.7
DAPOModel Size=7B, RL Algorithm=DAPO2026.0532.1
CSIPOModel Size=7B, RL Algorithm=CSIPO2026.0531.3
CSIPOModel Size=3B, RL Algorithm=CSIPO2026.0528.4
Agent 3 (Qwen2.5-1.5B-Instruct)Controller=Qwen2.5-7B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Qwen2.5-7B-Instruct, Agent 3=Qwen2.5-1.5B-Instruct2026.0527.3
AugHard@8Training regime=Rewritten AugHard@8 subset2026.0526.47
DAPOModel Size=3B, RL Algorithm=DAPO2026.0525.8
CSIPO + HistaModel Size=3B, RL Algorithm=CSIPO2026.0525.7
Full dataTraining regime=Full dataset2026.0525.37
RFGOOptimization=Reasoning Feature-Guided Optimization2026.0525
DAPO + HistaModel Size=3B, RL Algorithm=DAPO2026.0524.9
Controller V1Controller=Qwen2.5-7B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Qwen2.5-7B-Instruct, Agent 3=Qwen2.5-1.5B-Instruct2026.0524.7
AutoMixModel=Qwen2.5-1.5B2025.0923.2
DAREModel=Qwen2.5-Math-1.5B, Decoding=greedy decoding2026.0519.85
DAPO + HistaModel Size=1.5B, RL Algorithm=DAPO2026.0519.4
CSIPOModel Size=1.5B, RL Algorithm=CSIPO2026.0519.2
DAPOModel Size=1.5B, RL Algorithm=DAPO2026.0518.8
CSIPO + HistaModel Size=1.5B, RL Algorithm=CSIPO2026.0518
GSPO + HistaModel Size=1.5B, RL Algorithm=GSPO2026.0517.9
GSPOModel Size=1.5B, RL Algorithm=GSPO2026.0517.7
MoPPSModel=Qwen2.5-Math-1.5B, Decoding=greedy decoding2026.0516.54
hard@8Training regime=hard@8 subset2026.0515.81
DOTSModel=Qwen2.5-Math-1.5B, Decoding=greedy decoding2026.0515.44
GRPOGroup Size G=162026.0514.71
EDCOModel=Qwen2.5-Math-1.5B, Decoding=greedy decoding2026.0513.6
GRPOModel=Qwen2.5-Math-1.5B, Decoding=greedy decoding2026.0513.24
LLM-JudgeModel=Qwen2.5-Math-1.5B, Decoding=greedy decoding2026.0513.24
Previous FR EstimationModel=Qwen2.5-Math-1.5B, Decoding=greedy decoding2026.0513.24
GRPOGroup Size G=82026.0513.24
BPPOGroup Size G=82026.0513.24
BPPOGroup Size G=162026.0513.24
GRPOGroup Size G=322026.0512.5
BPPOGroup Size G=322026.0512.13
Agent 3 (Llama-3-2-1B-Instruct)Controller=Qwen3-4B-Base, Agent 1=Qwen3-30B-A3B-Instruct, Agent 2=Ministral-3-8B-Instruct, Agent 3=Llama-3-2-1B-Instruct2026.058.8
DAPO + HistaModel Size=0.5B, RL Algorithm=DAPO2026.056.4
DAPOModel Size=0.5B, RL Algorithm=DAPO2026.055.3