List your company
ResearchFormerly WizWand

Mathematics on AIME24

77.3AccuracyDSMoE
-2.228818.418139.06559.7119Jan 27, 2026Feb 16, 2026Mar 9, 2026Mar 30, 2026Apr 20, 2026May 11, 2026Jun 1, 2026
Updated 5mo ago

Evaluation results

MethodLinks
DSMoEModel=GPT-OSS-20B2026.0477.3——
Primitives-based MASBackbone=Qwen3-14B2026.0276.7—13.4
VotingBackbone=Qwen3-14B2026.0270—6.7
DSMoEModel=Qwen3-30B-Instruct2026.0470——
Primitives-based MASBackbone=Qwen3-4B2026.0266.7—23.4
LatentMASBackbone=Qwen3-14B2026.0266.7—3.4
ReviewBackbone=Qwen3-14B2026.0266.7—3.4
Qwen3-8BInference Mode=think2026.0366.67——
SSA-LLM-8BInference Mode=think2026.0366.67——
Qwen3.5Params=4B, Sampling mode=AR2026.0663.33——
VotingBackbone=Qwen3-4B2026.0263.3—20
SingleBackbone=Qwen3-14B2026.0263.3——
TextMASBackbone=Qwen3-14B2026.0263.3—0
PlanningBackbone=Qwen3-14B2026.0263.3—0
RICEModel=Qwen3-30B-Instruct2026.0463.3——
SFTModel=Qwen3-30B-Instruct2026.0463.3——
AdaRASCategory=Steering2026.0160.87——
FLAREParams=4B, Sampling mode=AR-Trust2026.0658.89——
LatentMASBackbone=Qwen3-4B2026.0256.7—13.4
ReviewBackbone=Qwen3-4B2026.0256.7—13.4
PlanningBackbone=Qwen3-4B2026.0256.7—13.4
Orig.Model=Qwen3-30B-Instruct2026.0456.7——
RICEModel=GPT-OSS-20B2026.0456.7——
SFTModel=GPT-OSS-20B2026.0456.7——
FLAREParams=4B, Sampling mode=Diffusion-Trust2026.0655.56——
Orig.Model=GPT-OSS-20B2026.0450——
CoTPrompting=Vanilla CoT, Base Model=Qwen3-1.7B2026.0147.83——
TextMASBackbone=Qwen3-4B2026.0246.7—3.4
ProbingCategory=Steering2026.0143.48——
SingleBackbone=Qwen3-4B2026.0243.3——
OpenReasoning-Nemotron-1.5BCategory=Post-training2026.0139.13——
DeepSeek-R1-Distill-Qwen-1.5BCategory=Post-training2026.0134.78——
Qwen3-8BInference Mode=no-think2026.0333.33——
FLAREParams=2B, Sampling mode=AR-Trust2026.0631.11——
FLAREParams=2B, Sampling mode=Diffusion-Trust2026.0631.11——
DARE + TIESGroup=Qwen3-4B-Base, Category=Merged Models2026.0630——
LLaDA2.1-minigeneration length=optimal, block length=optimal, denoising steps=optimal2026.0426.67——
General-ReasonerGroup=Qwen3-4B-Base, Category=Base and Expert Models2026.0626.67——
OpenThinker-3-1.5BCategory=Post-training2026.0126.08——
SSA-LLM-8BInference Mode=no-think2026.0323.33——
GRPOGroup=Qwen3-4B-Base, Category=Base and Expert Models2026.0623.33——
TIESGroup=Qwen3-4B-Base, Category=Merged Models2026.0623.33——
RAMGroup=Qwen3-4B-Base, Category=Merged Models2026.0623.33——
RESMERGEGroup=Qwen3-4B-Base, Category=Merged Models2026.0623.33——
TAGroup=Qwen3-4B-Base, Category=Merged Models2026.0620——
TSV-MergeGroup=Qwen3-4B-Base, Category=Merged Models2026.0620——
LLaDA2.0-minigeneration length=optimal, block length=optimal, denoising steps=optimal2026.0416.67——
SDAR-8B-Chatgeneration length=optimal, block length=optimal, denoising steps=optimal2026.0413.33——
SDAR-30B-A3Bgeneration length=optimal, block length=optimal, denoising steps=optimal2026.0413.33——
Qwen3-4BGroup=Qwen3-4B-Base, Category=Base and Expert Models2026.0613.33——
SpiralGroup=Qwen3-4B-Base, Category=Base and Expert Models2026.0613.33——
SDARParams=1.7B, Sampling mode=Diffusion2026.0610——
SDARParams=4B, Sampling mode=Diffusion2026.0610——
SDARParams=8B, Sampling mode=Diffusion2026.0610——
Qwen3.5Params=2B, Sampling mode=AR2026.068.89——
LLaDA-8B-Instructgeneration length=optimal, block length=optimal, denoising steps=optimal2026.042.08——
Dream-7B-Instructgeneration length=optimal, block length=optimal, denoising steps=optimal2026.040.83——
AMPOModel Category=Base and Expert Models2026.06—15—
BaseBackbone=olmo-3-7b-it-sft2026.04—7.5—
BaseBackbone=olmo-3.1-32b-it-sft2026.04—14.2—
DARE + TIESModel Category=Merged Models2026.06—16.67—
DPOBackbone=olmo-3-7b-it-sft2026.04—25.2—
DPOBackbone=olmo-3.1-32b-it-sft2026.04—32.1—
Falcon-H1R# Param.=7B2026.02—86.67—
FLARE-2BModel Size=2B, Training Protocol=FLARE, Decoding Strategy=AR-Trust sampling2026.06—31.11—
FLARE-4BModel Size=4B, Training Protocol=FLARE, Decoding Strategy=AR-Trust sampling2026.06—58.89—
FLARE-9BModel Size=9B, Training Protocol=FLARE, Decoding Strategy=AR-Trust sampling2026.06—63.33—
ISO-CModel Category=Merged Models2026.06—13.33—
ISO-CTSModel Category=Merged Models2026.06—16.67—
Mag+Backbone=olmo-3-7b-it-sft2026.04—29.6—
Mag+Backbone=olmo-3.1-32b-it-sft2026.04—40.8—
MemoryAgentModel Category=Base and Expert Models2026.06—8.67—
MiniCPM-4.1# Param.=8B2026.02—80.83—
MiniCPM-SALA# Param.=9B2026.02—83.75—
Ministral-3-R# Param.=8B2026.02—81.46—
MPO+Backbone=olmo-3-7b-it-sft2026.04—32.1—
MPO+Backbone=olmo-3.1-32b-it-sft2026.04—46—
Nemotron-Nano-v2# Param.=9B2026.02—71.67—
Pair+Backbone=olmo-3-7b-it-sft2026.04—32.1—
Pair+Backbone=olmo-3.1-32b-it-sft2026.04—46.7—
Qwen2.5-7B-InstructModel Category=Base and Expert Models2026.06—13—
Qwen3# Param.=8B2026.02—73.33—
Qwen3.5-2B (released)Model Size=2B, Training Protocol=Released, Decoding Strategy=Native AR decoding2026.06—8.89—
Qwen3.5-2B + AR-SFTModel Size=2B, Training Protocol=AR-SFT, Decoding Strategy=Native AR decoding2026.06—27.78—
Qwen3.5-4B (released)Model Size=4B, Training Protocol=Released, Decoding Strategy=Native AR decoding2026.06—63.33—
Qwen3.5-4B + AR-SFTModel Size=4B, Training Protocol=AR-SFT, Decoding Strategy=Native AR decoding2026.06—62.22—
Qwen3.5-9B (released)Model Size=9B, Training Protocol=Released, Decoding Strategy=Native AR decoding2026.06—65.56—
Qwen3.5-9B + AR-SFTModel Size=9B, Training Protocol=AR-SFT, Decoding Strategy=Native AR decoding2026.06—66.67—
RAMModel Category=Merged Models2026.06—13—
RESMERGEModel Category=Merged Models2026.06—16.67—
RFTBackbone=olmo-3-7b-it-sft2026.04—15.8—
RFTBackbone=olmo-3.1-32b-it-sft2026.04—27.1—
Soft+Backbone=olmo-3-7b-it-sft2026.04—31.9—
Soft+Backbone=olmo-3.1-32b-it-sft2026.04—44.2—
TAModel Category=Merged Models2026.06—13.33—
TIESModel Category=Merged Models2026.06—15.33—
ToolRLModel Category=Base and Expert Models2026.06—13.67—
TSV-MergeModel Category=Merged Models2026.06—12.67—