List your company
ResearchFormerly WizWand

Mathematical Reasoning on HMMT25 (Accuracy (%))

92.5Accuracy (%)TRICE-30B
-0.99623.27747.5571.823Oct 7, 2025Nov 12, 2025Dec 18, 2025Jan 24, 2026Mar 1, 2026Apr 6, 2026May 13, 2026
Updated 5mo ago

Evaluation results

MethodLinks
TRICE-30BTool Use=Yes2026.0592.5
DeepSeek-V3.2-ThinkingTool Use=No2026.0590.8
Qwen3-235B-A22B-ThinkingTool Use=No2026.0588.8
GSPO + IBPO(mask)Model=Qwen3-Next2026.0481.1
GSPO + IBPOModel=Qwen3-Next2026.0480.4
GSPO + IBPO(ratio)Model=Qwen3-Next2026.0480
GSPO with Best-of-NModel=Qwen3-Next, N=82026.0477.2
GSPO with promptModel=Qwen3-Next2026.0476.5
GSPO with SCoReModel=Qwen3-Next2026.0476.3
GSPOModel=Qwen3-Next2026.0475.9
ParaManagerOrchestration Strategy=Unified2026.0463.33
GSPO + IBPO(mask)Model=Qwen3-32B2026.0463.1
GSPO + IBPOModel=Qwen3-32B2026.0462.6
GSPO + IBPO(ratio)Model=Qwen3-32B2026.0462.2
ParaManager-SerialOrchestration Strategy=Serial2026.0460
ToolOrchestraOrchestration Strategy=Serial Orchestration2026.0458.33
GSPO with Best-of-NModel=Qwen3-32B, N=82026.0457.1
PuppeteerOrchestration Strategy=Serial Orchestration2026.0457.08
GSPO with SCoReModel=Qwen3-32B2026.0456.7
GPT-OSS-20BOrchestration Strategy=Base+Tool2026.0456.67
GPT-OSS-20BOrchestration Strategy=Majority Vote2026.0456.67
ParaManager-SFTOrchestration Strategy=SFT2026.0456.67
GSPO with promptModel=Qwen3-32B2026.0456.4
GSPOModel=Qwen3-32B2026.0455.6
CL-EGRSDBackbone=Qwen3-8B2026.0552.5
GPT-OSS-20BOrchestration Strategy=Self Refine2026.0452.08
EvoFlowOrchestration Strategy=Static Workflow2026.0452.08
ParaManager-MonoOrchestration Strategy=Mono2026.0450
KnowRL-Nemotron-1.5BHint Setting=CSS, Evaluation Protocol=mean@322026.0448.75
RouterOrchestration Strategy=Static Workflow2026.0447.92
GPT-OSS-20BOrchestration Strategy=Base2026.0447.5
SFTBackbone=Qwen3-8B2026.0546.67
OPSDBackbone=Qwen3-8B2026.0545.84
EGRSDBackbone=Qwen3-8B2026.0545.42
Baseline (no-train)Backbone=Qwen3-8B2026.0545.1
KnowRL-Nemotron-1.5BHint Setting=CBRS, Evaluation Protocol=mean@322026.0445
QuestAHint Setting=CSS, Evaluation Protocol=mean@322026.0444.35
Meta Agent SearchOrchestration Strategy=Static Workflow2026.0444.17
QuestAHint Setting=CBRS, Evaluation Protocol=mean@322026.0443.78
Qwen3-30B A3B-Instruct-2507Orchestration Strategy=Self Refine2026.0443.75
Qwen3-30B A3B-Instruct-2507Orchestration Strategy=Base+Tool2026.0443.33
Qwen3-30B A3B-Instruct-2507Orchestration Strategy=Majority Vote2026.0443.33
CRISPBackbone=Qwen3-8B2026.0542.92
Qwen3-30B A3B-Instruct-2507Orchestration Strategy=Base2026.0442.5
JustRLHint Setting=CBRS, Evaluation Protocol=mean@322026.0441.81
GRPOBackbone=Qwen3-8B2026.0541.67
JustRLHint Setting=CSS, Evaluation Protocol=mean@322026.0441.54
KnowRL-Nemotron-1.5BHint Setting=w/o KP, Evaluation Protocol=mean@322026.0441.04
QuestAHint Setting=w/o KP, Evaluation Protocol=mean@322026.0440.94
JustRLHint Setting=w/o KP, Evaluation Protocol=mean@322026.0440.63
Qwen3-4B Instruct-2507Orchestration Strategy=Majority Vote2026.0440
Qwen3-4B Instruct-2507Orchestration Strategy=Self Refine2026.0437.5
Nemotron-1.5BHint Setting=CSS, Evaluation Protocol=mean@322026.0435.77
Nemotron-1.5BHint Setting=CBRS, Evaluation Protocol=mean@322026.0434.79
Qwen3-4B Instruct-2507Orchestration Strategy=Base2026.0434.17
Qwen3-4B Instruct-2507Orchestration Strategy=Base+Tool2026.0434.17
Nemotron-1.5BHint Setting=w/o KP, Evaluation Protocol=mean@322026.0430.63
Qwen3-Coder 30B-A3B-InstructOrchestration Strategy=Base+Tool2026.0423.33
ARM + MPFoundation Model=GPT-4.1-nano2025.1022.4
DGO (TTS)Model=Qwen3-14B-Base, Inference Mode=Test-Time Scaling2026.0322.29
LLM-DebateFoundation Model=GPT-4.1-nano2025.1016.7
DGO (TTS)Model=Qwen3-4B-Base, Inference Mode=Test-Time Scaling2026.0315.83
DGO (zero)Model=Qwen3-14B-Base, Inference Mode=Intrinsic Inference2026.0315.21
ARMFoundation Model=GPT-4.1-nano2025.1014.6
DGO (TTS)Model=Qwen3-8B-Base, Inference Mode=Test-Time Scaling2026.0314.37
GRPOModel=Qwen3-14B-Base2026.0314.37
DAPOModel=Qwen3-14B-Base2026.0314.17
CoT-SCFoundation Model=GPT-4.1-nano2025.1013.5
Qwen3-Coder 30B-A3B-InstructOrchestration Strategy=Majority Vote2026.0413.33
Qwen3-Coder 30B-A3B-InstructOrchestration Strategy=Self Refine2026.0412.5
AFlowFoundation Model=GPT-4.1-nano, Evaluation Setup=test set2025.1012
DGO (zero)Model=Qwen3-8B-Base, Inference Mode=Intrinsic Inference2026.0311.88
DGO (zero)Model=Qwen3-4B-Base, Inference Mode=Intrinsic Inference2026.0311.67
Qwen3-Coder 30B-A3B-InstructOrchestration Strategy=Base2026.0410.83
AFlowFoundation Model=GPT-4.1-nano, Evaluation Setup=1000-sample2025.1010.4
CoTFoundation Model=GPT-4.1-nano2025.109.9
Self-RefineFoundation Model=GPT-4.1-nano2025.109.4
ARM + MPFoundation Model=GPT-4o2025.109.4
DAPOModel=Qwen3-4B-Base2026.039.17
GRPOModel=Qwen3-8B-Base2026.039.17
DAPOModel=Qwen3-8B-Base2026.039.17
GRPOModel=Qwen3-4B-Base2026.038.75
ConSPOBackbone=Qwen3-4B-Base2026.058
ADASFoundation Model=GPT-4.1-nano, Evaluation Setup=1000-sample2025.106.8
ARM + MPFoundation Model=LLaMA-3.3-70B2025.106.8
SFTModel=Qwen3-14B-Base2026.036.46
ARMFoundation Model=GPT-4o2025.105.7
CoT-SCFoundation Model=LLaMA-3.3-70B2025.105.7
DisCOBackbone=Qwen3-4B-Base2026.055.3
ADASFoundation Model=GPT-4.1-nano, Evaluation Setup=test set2025.105.2
ARMFoundation Model=LLaMA-3.3-70B2025.105.2
Self-RefineFoundation Model=LLaMA-3.3-70B2025.104.2
LLM-DebateFoundation Model=LLaMA-3.3-70B2025.104.2
AFlowFoundation Model=GPT-4o, Evaluation Setup=test set2025.103.6
SFTModel=Qwen3-8B-Base2026.033.54
LLM-DebateFoundation Model=GPT-4o2025.103.1
CoTFoundation Model=LLaMA-3.3-70B2025.103.1
AFlowFoundation Model=LLaMA-3.3-70B, Evaluation Setup=1000-sample2025.103.1
GRPOBackbone=Qwen3-4B-Base2026.052.6
Dr. GRPOBackbone=Qwen3-4B-Base2026.052.6