List your company
ResearchFormerly WizWand

Science Reasoning on ARC-C

97AccuracyQwen3-Next-80B + THINKBRAKE
48.53661.11873.786.282Oct 1, 2025Nov 12, 2025Dec 25, 2025Feb 6, 2026Mar 21, 2026May 3, 2026Jun 15, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Qwen3-Next-80B + THINKBRAKEBackbone=Qwen3-Next-80B, Decoding Strategy=THINKBRAKE2025.1097-21.8—
Qwen3-Next-80BBackbone=Qwen3-Next-80B, Decoding Strategy=Base2025.1096.8——
Qwen3-14B + NoWaitBackbone=Qwen3-14B, Decoding Strategy=NoWait2025.1096.2-8.1—
NoWaitBase Model=Qwen3-14B2025.1096.2-8.1—
Qwen3-14BBackbone=Qwen3-14B, Decoding Strategy=Base2025.1095.9——
Qwen3-14BStrategy=Baseline2025.1095.9——
Qwen3-14B + THINKBRAKEBackbone=Qwen3-14B, Decoding Strategy=THINKBRAKE2025.1095.2-18.2—
THINKBRAKEBase Model=Qwen3-14B2025.1095.2-18.2—
Qwen3-32B + ThinkLessBackbone=Qwen3-32B, Decoding Strategy=ThinkLess2025.1095.1-100—
Qwen3-14B + THINKBRAKE-pBackbone=Qwen3-14B, Decoding Strategy=THINKBRAKE-p2025.1094.7-17.9—
Qwen3-4B-Thinking + Dynasor-CoTBackbone=Qwen3-4B-Thinking, Decoding Strategy=Dynasor-CoT2025.1094.5-39.1—
Dynasor-CoTBase Model=Qwen3-4B-Thinking2025.1094.5-39.1—
Qwen3-4B-ThinkingBackbone=Qwen3-4B-Thinking, Decoding Strategy=Base2025.1094.3——
Qwen3-14B + ThinkLessBackbone=Qwen3-14B, Decoding Strategy=ThinkLess2025.1094.3-100—
Qwen3-4B-ThinkingStrategy=Baseline2025.1094.3——
Qwen3-4B-Thinking + ThinkLessBackbone=Qwen3-4B-Thinking, Decoding Strategy=ThinkLess2025.1094.2-100—
ThinkLessBase Model=Qwen3-4B-Thinking2025.1094.2-100—
Qwen3-4B-Thinking + THINKBRAKEBackbone=Qwen3-4B-Thinking, Decoding Strategy=THINKBRAKE2025.1094-19.9—
Qwen3-4BBackbone=Qwen3-4B, Decoding Strategy=Base2025.1094——
THINKBRAKEBase Model=Qwen3-4B-Thinking2025.1094-19.9—
Qwen3-4BStrategy=Baseline2025.1094——
Qwen3-4B + THINKBRAKEBackbone=Qwen3-4B, Decoding Strategy=THINKBRAKE2025.1093.9-11.5—
THINKBRAKEBase Model=Qwen3-4B2025.1093.9-11.5—
Qwen3-4B-Thinking + THINKBRAKE-pBackbone=Qwen3-4B-Thinking, Decoding Strategy=THINKBRAKE-p2025.1093.7-31.7—
THINKBRAKE-pBase Model=Qwen3-4B-Thinking2025.1093.7-31.7—
Qwen3-4B + NoWaitBackbone=Qwen3-4B, Decoding Strategy=NoWait2025.1093.4-14.4—
Qwen3-4B + THINKBRAKE-pBackbone=Qwen3-4B, Decoding Strategy=THINKBRAKE-p2025.1093.4-15—
NoWaitBase Model=Qwen3-4B2025.1093.4-14.4—
Qwen3-4B-Thinking + DEERBackbone=Qwen3-4B-Thinking, Decoding Strategy=DEER2025.1092.8-9—
DEERBase Model=Qwen3-4B-Thinking2025.1092.8-9—
Qwen3-4B-Thinking + NoWaitBackbone=Qwen3-4B-Thinking, Decoding Strategy=NoWait2025.1092.714.6—
NoWaitBase Model=Qwen3-4B-Thinking2025.1092.714.6—
Qwen3-32B + THINKBRAKE-pBackbone=Qwen3-32B, Decoding Strategy=THINKBRAKE-p2025.1092.2-11.7—
Qwen3-32BBackbone=Qwen3-32B, Decoding Strategy=Base2025.1092.1——
Qwen3-32BStrategy=Baseline2025.1092.1——
Qwen3-32B + THINKBRAKEBackbone=Qwen3-32B, Decoding Strategy=THINKBRAKE2025.1091-8—
THINKBRAKEBase Model=Qwen3-32B2025.1091-8—
Qwen3-32B + NoWaitBackbone=Qwen3-32B, Decoding Strategy=NoWait2025.1090.19—
NoWaitBase Model=Qwen3-32B2025.1090.19—
Qwen3-4B + ThinkLessBackbone=Qwen3-4B, Decoding Strategy=ThinkLess2025.1089.3-100—
LLaDA + DreamEnsemble Type=Intermediate-generation Ensemble, Scoring Function=Token change count2026.0689.16——
LLaDA + DreamEnsemble Type=Intermediate-generation Ensemble, Scoring Function=Top-1 probability2026.0688.82——
LLaDA + DreamEnsemble Type=Intermediate-generation Ensemble, Scoring Function=Probability margin2026.0688.74——
LLaDA + DreamEnsemble Type=Post-generation Ensemble2026.0688.57——
LLaDA + DreamEnsemble Type=Intermediate-generation Ensemble, Scoring Function=Entropy2026.0688.57——
DreamEnsemble Type=Single Model2026.0686.69——
LLaDAEnsemble Type=Single Model2026.0685.15——
Phi-4-Reasoning + THINKBRAKE-pBackbone=Phi-4-Reasoning, Decoding Strategy=THINKBRAKE-p2025.1084.2-49.2—
Phi-4-Reasoning + ThinkLessBackbone=Phi-4-Reasoning, Decoding Strategy=ThinkLess2025.1082.7-100—
Phi-4-ReasoningBackbone=Phi-4-Reasoning, Decoding Strategy=Base2025.1080——
Phi-4-Reasoning + THINKBRAKEBackbone=Phi-4-Reasoning, Decoding Strategy=THINKBRAKE2025.1080-25.3—
Phi-4-ReasoningStrategy=Baseline2025.1080——
THINKBRAKEBase Model=Phi-4-Reasoning2025.1080-25.3—
DeepSeek-R1-7B + Dynasor-CoTBackbone=DeepSeek-R1-7B, Decoding Strategy=Dynasor-CoT2025.1077.8-53.6—
Phi-4-Reasoning + NoWaitBackbone=Phi-4-Reasoning, Decoding Strategy=NoWait2025.1077.6-7—
NoWaitBase Model=Phi-4-Reasoning2025.1077.6-7—
DeepSeek-R1-7B + DEERBackbone=DeepSeek-R1-7B, Decoding Strategy=DEER2025.1073.9-10.6—
DeepSeek-R1-7B + THINKBRAKEBackbone=DeepSeek-R1-7B, Decoding Strategy=THINKBRAKE2025.1070.7-35.3—
THINKBRAKEBase Model=DeepSeek-R1-7B2025.1070.7-35.3—
DeepSeek-R1-7BBackbone=DeepSeek-R1-7B, Decoding Strategy=Base2025.1067.7——
DeepSeek-R1-7BStrategy=Baseline2025.1067.7——
DeepSeek-R1-7B + THINKBRAKE-pBackbone=DeepSeek-R1-7B, Decoding Strategy=THINKBRAKE-p2025.1067.2-31—
DeepSeek-R1-7B + NoWaitBackbone=DeepSeek-R1-7B, Decoding Strategy=NoWait2025.1064.6-12.6—
NoWaitBase Model=DeepSeek-R1-7B2025.1064.6-12.6—
DeepSeek-R1-7B + ThinkLessBackbone=DeepSeek-R1-7B, Decoding Strategy=ThinkLess2025.1050.4-100—
DAPOBase Model=Qwen3-4B-Instruct-25072026.05——88.5
DAPOBase Model=Phi-4-mini-instruct2026.05——85
Dr.GRPOBase Model=Qwen3-4B-Instruct-25072026.05——88.3
Dr.GRPOBase Model=Phi-4-mini-instruct2026.05——84.8
GRPOBase Model=Qwen3-4B-Instruct-25072026.05——88
GRPOBase Model=Phi-4-mini-instruct2026.05——84.6
SDPOBase Model=Qwen3-4B-Instruct-25072026.05——89
SDPOBase Model=Phi-4-mini-instruct2026.05——85.4
Self-OPPOBase Model=Qwen3-4B-Instruct-2507, Optimization Framework=GRPO2026.05——89.5
Self-OPPOBase Model=Qwen3-4B-Instruct-2507, Optimization Framework=DAPO2026.05——89.6
Self-OPPOBase Model=Phi-4-mini-instruct, Optimization Framework=GRPO2026.05——86.2
Self-OPPOBase Model=Phi-4-mini-instruct, Optimization Framework=DAPO2026.05——86.4
SFTBase Model=Qwen3-4B-Instruct-25072026.05——87.1
SFTBase Model=Phi-4-mini-instruct2026.05——83.5
Teacher-OPPOBase Model=Qwen3-4B-Instruct-2507, Optimization Framework=GRPO2026.05——89.8
Teacher-OPPOBase Model=Qwen3-4B-Instruct-2507, Optimization Framework=DAPO2026.05——90
Teacher-OPPOBase Model=Phi-4-mini-instruct, Optimization Framework=GRPO2026.05——86.8
Teacher-OPPOBase Model=Phi-4-mini-instruct, Optimization Framework=DAPO2026.05——87