List your company

Mathematical Reasoning on Minerva (Avg@16)

46.78Avg@16DISPO
16.328824.234432.1440.0456Feb 1, 2026Feb 19, 2026Mar 9, 2026Mar 27, 2026Apr 14, 2026May 2, 2026May 21, 2026

Evaluation results

MethodLinks
DISPOModel=Qwen3-14B-Base2026.0246.78
CISPOModel=Qwen3-14B-Base2026.0245.66
DAPOModel=Qwen3-14B-Base2026.0245.22
DAPOModel=Qwen3-8B-Base2026.0244.85
DISPOModel=Qwen3-30B-A3B-Base (MoE)2026.0244.6
DISPOModel=Qwen3-8B-Base2026.0244.6
CISPOModel=Qwen3-30B-A3B-Base (MoE)2026.0244.26
CISPOModel=Qwen3-8B-Base2026.0243.52
DAPOModel=Qwen3-30B-A3B-Base (MoE)2026.0243.45
Starting checkpointModel=Qwen3-14B-Base2026.0236.1
S-trace-0.9Backbone=Qwen3-4B2026.0533.28
DASDModel Scale=8B2026.0533
DASDModel Scale=4B2026.0531
Starting checkpointModel=Qwen3-8B-Base2026.0230.95
GRPO(λ)-0.9Backbone=Qwen3-4B2026.0530.52
Starting checkpointModel=Qwen3-30B-A3B-Base (MoE)2026.0230.35
GRPOBackbone=Qwen3-4B2026.0530.02
RLSDModel Scale=8B2026.0529.7
HEPOModel Scale=8B2026.0528.7
OPOBackbone=Qwen3-4B2026.0528.5
RLSDModel Scale=4B2026.0528.5
GRPOModel Scale=4B2026.0528.2
P-trace-0.9Backbone=Qwen3-4B2026.0528.18
GRPO(λ)-0.99Backbone=Qwen3-4B2026.0527.94
GRPOModel Scale=8B2026.0527.7
HEPOModel Scale=4B2026.0527.1
SRPOModel Scale=4B2026.0526.1
SRPOModel Scale=8B2026.0525.9
DASDModel Scale=1.7B2026.0525.3
RLSDModel Scale=1.7B2026.0525.1
OPSDModel Scale=4B2026.0524.8
Qwen3Model Scale=8B2026.0524.7
Qwen3Model Scale=4B2026.0524.6
SDPOModel Scale=4B2026.0524.4
SDPOModel Scale=8B2026.0524.1
OPSDModel Scale=8B2026.0523.8
HEPOModel Scale=1.7B2026.0522.8
GRPOModel Scale=1.7B2026.0522.6
SRPOModel Scale=1.7B2026.0520.1
SDPOModel Scale=1.7B2026.0517.8
Qwen3Model Scale=1.7B2026.0517.5
OPSDModel Scale=1.7B2026.0517.5