List your company

Mathematical Reasoning on AIME 2026 (accuracy)

93.3AIME 2026 AccuracySU-01
-1.02823.46147.9572.439Apr 28, 2026May 4, 2026May 11, 2026May 17, 2026May 24, 2026May 30, 2026Jun 6, 2026

Evaluation results

MethodLinks
SU-01Number of runs=82026.0593.3
Qwen3.6-35B-A3BNumber of runs=82026.0592.9
Gemma-4-31BNumber of runs=82026.0591.3
Nemotron-Cascade-2Number of runs=82026.0590
P1-30B-A3BNumber of runs=82026.0589.6
GLM-4.7-FlashNumber of runs=82026.0588.3
RecursiveMASrecursion round=32026.0486.7
SeirênesBackbone=Qwen3-30B-A3B, same budget=true2026.0579.7
Single Agenttraining protocol=Full-SFT2026.0476.7
TextGradrecursion round=32026.0476.7
FullModel=Qwen3-14B, KV Cache Budget=100%2026.0676.04
DAPOBackbone=Qwen3-30B-A3B, same budget=true2026.0575.2
Qwen3-30B-A3BBackbone=Qwen3-30B-A3B2026.0575.1
Single Agenttraining protocol=LoRA2026.0473.3
Recursive-TextMASrecursion round=32026.0473.3
VASE-ATTNVModel=Qwen3-14B, KV Cache Budget=~25%2026.0672.29
SeirênesBackbone=Qwen3-4B-Instruct, same budget=true2026.0570.8
VASE-DKVModel=Qwen3-14B, KV Cache Budget=~25%2026.0668.75
SeerAttention-RModel=Qwen3-14B, KV Cache Budget=~25%2026.0665.62
Dr.GRPOBackbone=Qwen3-4B-Instruct, same budget=true2026.0564.9
InTBackbone=Qwen3-4B-Instruct2026.0564.7
Mixture-of-Agents (MoA)recursion round=32026.0463.3
LoopLMrecursion round=32026.0463.3
FullModel=Qwen3-4B, KV Cache Budget=100%2026.0662.71
DAPOBackbone=Qwen3-4B-Instruct, same budget=true2026.0562.2
VASE-ATTNVModel=Qwen3-4B, KV Cache Budget=~25%2026.0662.08
VASE-DKVModel=Qwen3-4B, KV Cache Budget=~25%2026.0661.88
CurDKVModel=Qwen3-14B, KV Cache Budget=~25%2026.0661.25
SeerAttention-RModel=Qwen3-4B, KV Cache Budget=~25%2026.0660.42
R-KVModel=Qwen3-14B, KV Cache Budget=~25%2026.0660.21
SAGEBackbone=Qwen3-4B-Instruct2026.0557.8
Qwen3-4B-InstructBackbone=Qwen3-4B-Instruct2026.0556.8
R-KVModel=Qwen3-4B, KV Cache Budget=~25%2026.0654.58
CurDKVModel=Qwen3-4B, KV Cache Budget=~25%2026.0654.58
Scaf-GRPOBackbone=Qwen3-4B-Instruct2026.0554.3
SnapKVModel=Qwen3-4B, KV Cache Budget=~25%2026.0653.33
DAPO w/ EDASBackbone=Qwen3-8B2026.0548.65
RREDCoTGeneration Length=25k tokens, Training Dataset=Numina-CoT2026.0647.5
CIKAParameter Count=7B, Status=Frozen2026.0546.7
DelTABackbone=Qwen3-14B-Base, Fine-tuning (FT)=false2026.0545.21
DAPO w/ EDASBackbone=Qwen3-4B2026.0544.27
GRPOGeneration Length=25k tokens, Training Dataset=Numina-CoT2026.0644.2
FIPOBackbone=Qwen3-14B-Base, Fine-tuning (FT)=false2026.0542.5
DAPOBackbone=Qwen3-14B-Base, Fine-tuning (FT)=true2026.0541.46
SAPOBackbone=Qwen3-14B-Base, Fine-tuning (FT)=false2026.0541.46
DAPOBackbone=Qwen3-4B2026.0540.73
DAPOBackbone=Qwen3-14B-Base, Fine-tuning (FT)=false2026.0539.79
DAPOBackbone=Qwen3-8B2026.0539.17
DAPO w/ En AdvBackbone=Qwen3-8B2026.0533.23
DelTABackbone=Qwen3-8B-Base, Fine-tuning (FT)=false2026.0528.12
DAPO w/ En AdvBackbone=Qwen3-4B2026.0526.77
Qwen3-4B Instruct 2507 (starting point)Generation Length=25k tokens, Training Dataset=Numina-CoT2026.0626.7
DAPOBackbone=Qwen3-8B-Base, Fine-tuning (FT)=true2026.0526.46
SAPOBackbone=Qwen3-8B-Base, Fine-tuning (FT)=false2026.0526.25
DAPOBackbone=Qwen3-8B-Base, Fine-tuning (FT)=false2026.0524.17
FIPOBackbone=Qwen3-8B-Base, Fine-tuning (FT)=false2026.0523.96
ConSteer-RLBackbone=Qwen3-8B-Base2026.0623.8
DAPO w/ EDASBackbone=Qwen3-4B-Base2026.0520.97
GRPOBackbone=Qwen3-8B-Base2026.0620.7
ConSteer-RLBackbone=Qwen3-4B-Base2026.0620.2
GRPO w/ EDASBackbone=Qwen3-8B2026.0519.06
PKPOBackbone=Qwen3-8B2026.0518.85
GRPOBackbone=Qwen3-8B2026.0518.43
BaselineBackbone=Qwen3-4B2026.0518.02
SeirênesBackbone=Qwen2.5-7B-Instruct, same budget=true2026.0517.8
ConSteer-RLBackbone=Qwen2.5-Math-7B2026.0616.7
BaselineBackbone=Qwen3-8B2026.0515.1
GRPOBackbone=Qwen3-4B-Base2026.0614
DAPO w/ En AdvBackbone=Qwen3-4B-Base2026.0512.71
LUFFYBackbone=Qwen2.5-7B-Instruct2026.0512.3
DAPOBackbone=Qwen3-4B-Base2026.0510.73
BaselineBackbone=Qwen3-4B-Base2026.0610.7
GRPOBackbone=Qwen2.5-Math-7B2026.0610
SAGEBackbone=Qwen2.5-7B-Instruct2026.059.3
DAPOBackbone=Qwen2.5-7B-Instruct, same budget=true2026.059.3
ExGRPOBackbone=Qwen2.5-7B-Instruct2026.059
BaselineBackbone=Qwen2.5-Math-7B2026.066.7
BaselineBackbone=Qwen3-8B-Base2026.066.2
Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct2026.055.3
BaselineBackbone=Qwen3-4B-Base2026.052.6