List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 2025

95AccuracyGemini 3-Pro
48.260.3572.584.65Jan 31, 2025Apr 3, 2025Jun 5, 2025Aug 7, 2025Oct 9, 2025Dec 11, 2025Feb 12, 2026
Updated 8mo ago

Evaluation results

MethodLinks
Gemini 3-Pro2026.0295
GPT-5 (High)2026.0294.6
GPT-5.12026.0294
PRISMBase Model=Qwen3-30B-A3B, Zero-shot prompting=true, Thinking mode=false, Proposers=3, Reviewer=1, Synthesis iterations=32026.0293.3
DMind-3Number of Parameters=21B2026.0293.3
DS V3.2-Thinking2026.0293.1
Agentic Proposingtraining_budget=11,000 mixed trajectories, recipe=GRPO, evaluation=Mean@64, proposer=Agentic-Proposer-30B2026.0291.6
Interfaze-Beta2026.0290
PromptCoT 2.0training_budget=11,000 mixed trajectories, recipe=GRPO, evaluation=Mean@642026.0289.8
DeepSeek V3.22026.0289.3
ERNIE 5.02026.0289.06
Gemini 2.5-Pro2026.0288
Gemini 2.5 Pro2026.0287.7
Claude Sonnet 4.52026.0287
MoABase Model=Qwen3-30B-A3B, Zero-shot prompting=true, Thinking mode=false, Agents per layer=32026.0286.7
Gemini 2.5 proEvaluation Protocol=pass@12025.0986.7
OpenMathReasoningtraining_budget=11,000 mixed trajectories, recipe=GRPO, evaluation=Mean@642026.0286.1
Qwen3-30B-A3B-Thinking-2507mode=zero-shot, evaluation=Mean@642026.0285
OpenR1training_budget=11,000 mixed trajectories, recipe=GRPO, evaluation=Mean@642026.0284.9
OpenThoughts-S3training_budget=11,000 mixed trajectories, recipe=GRPO, evaluation=Mean@642026.0284.7
DMind-3-miniNumber of Parameters=4B2026.0283.3
OpenCodeReasoningtraining_budget=11,000 mixed trajectories, recipe=GRPO, evaluation=Mean@642026.0281.1
MiniMax M2.12026.0281
o3-mini-highaccess=API only, examples_for_finetuning=N/A2025.0180.9
Qwen3-235B-A22BModel Category=Large Language Models, Model Scale=235B-A22B, Reasoning Strategy=Thinking2025.1280.78
Two HeadsBase Model=Qwen3-30B-A3B, Zero-shot prompting=true, Thinking mode=false, Solvers=K=32026.0280
Majority Voting (Shortest-K)Model=Qwen3-14B, K=8, Strategy=Shortest-K Merge2025.1280
FIGR2025.1279.32
THINKMERGE Shortest-K Merge (D)Model=Qwen3-14B, K=8, Strategy=Shortest-K Merge2025.1278.7
Majority Voting (Shortest-K)Model=Qwen3-14B, K=4, Strategy=Shortest-K Merge2025.1278
THINKMERGE DirectMerge (A)Model=Qwen3-14B, K=8, Strategy=All-Reduce2025.1278
THINKMERGE Early-Ready (C)Model=Qwen3-14B, K=8, Strategy=Early-Ready2025.1278
Claude Opus 4.1Evaluation Protocol=pass@12025.0978
Majority Voting (Early-Ready)Model=Qwen3-14B, K=8, Strategy=Early-Ready2025.1277.4
DeepSeek-V3.2Type=Single-model reference, Zero-shot prompting=true, Extended reasoning (thinking mode)=true2026.0276.8
Majority Voting (Early-Ready)Model=Qwen3-14B, K=4, Strategy=Early-Ready2025.1276
THINKMERGE Shortest-K Merge (D)Model=Qwen3-14B, K=4, Strategy=Shortest-K Merge2025.1276
Majority Voting (Shortest-K)Model=Qwen3-14B, K=2, Strategy=Shortest-K Merge2025.1275.3
Majority Voting (Shortest-K)Model=Qwen3-4B, K=4, Strategy=Shortest-K Merge2025.1275.3
Majority Voting (Shortest-K)Model=Qwen3-4B, K=8, Strategy=Shortest-K Merge2025.1275.3
THINKMERGE Shortest-K Merge (D)Model=Qwen3-14B, K=2, Strategy=Shortest-K Merge2025.1274.7
Claude Sonnet 4Reasoning=Thinking2026.0274.3
Majority VotingModel=Qwen3-14B, K=8, Strategy=All-Reduce2025.1274
Qwen3-235B-A22BType=Single-model reference, Zero-shot prompting=true, Thinking mode=false2026.0273.3
Claude Opus 4Reasoning=Thinking2026.0273.3
THINKMERGE Early-Ready (C)Model=Qwen3-14B, K=2, Strategy=Early-Ready2025.1273.3
Majority VotingModel=Qwen3-14B, K=4, Strategy=All-Reduce2025.1273.3
THINKMERGE Trimming (B)Model=Qwen3-14B, K=4, Strategy=All-Reduce2025.1273.3
THINKMERGE Early-Ready (C)Model=Qwen3-14B, K=4, Strategy=Early-Ready2025.1273.3
Majority Voting (Early-Ready)Model=Qwen3-4B, K=8, Strategy=Early-Ready2025.1273.3
THINKMERGE Trimming (B)Model=Qwen3-14B, K=8, Strategy=All-Reduce2025.1273.3
Qwen3-32BModel Category=Large Language Models, Model Scale=32B, Reasoning Strategy=Thinking2025.1272.9
THINKMERGE Trimming (B)Model=Qwen3-14B, K=2, Strategy=All-Reduce2025.1272.7
Majority Voting (Early-Ready)Model=Qwen3-4B, K=4, Strategy=Early-Ready2025.1272.7
THINKMERGE Early-Ready (C)Model=Qwen3-4B, K=4, Strategy=Early-Ready2025.1272.7
THINKMERGE Shortest-K Merge (D)Model=Qwen3-4B, K=8, Strategy=Shortest-K Merge2025.1272.7
THINKMERGE DirectMerge (A)Model=Qwen3-14B, K=2, Strategy=All-Reduce2025.1272
Majority Voting (Early-Ready)Model=Qwen3-14B, K=2, Strategy=Early-Ready2025.1272
THINKMERGE DirectMerge (A)Model=Qwen3-4B, K=4, Strategy=All-Reduce2025.1272
THINKMERGE DirectMerge (A)Model=Qwen3-14B, K=4, Strategy=All-Reduce2025.1272
THINKMERGE Early-Ready (C)Model=Qwen3-4B, K=8, Strategy=Early-Ready2025.1270.8
Majority Voting (Shortest-K)Model=Qwen3-4B, K=2, Strategy=Shortest-K Merge2025.1270.7
Majority VotingModel=Qwen3-14B, K=2, Strategy=All-Reduce2025.1270.7
o3-mini-mediumaccess=API only, examples_for_finetuning=N/A2025.0170.4
r1access=Open Weights, examples_for_finetuning=>800K2025.0170
ReConcileBase Model=Qwen3-30B-A3B, Zero-shot prompting=true, Thinking mode=false, Agents=3, Discussion rounds=22026.0270
Qwen3-30B-A3BType=Single-model reference, Zero-shot prompting=true, Thinking mode=false2026.0270
R1# Examples=>>800K2025.1270
THINKMERGE DirectMerge (A)Model=Qwen3-4B, K=8, Strategy=All-Reduce2025.1270
THINKMERGE Shortest-K Merge (D)Model=Qwen3-4B, K=4, Strategy=Shortest-K Merge2025.1269.3
Text-only RLMode=Text-only, Reasoning Strategy=RL2025.1269.22
Majority VotingModel=Qwen3-4B, K=8, Strategy=All-Reduce2025.1268.7
THINKMERGE Trimming (B)Model=Qwen3-4B, K=8, Strategy=All-Reduce2025.1268.7
Majority VotingModel=Qwen3-4B, K=4, Strategy=All-Reduce2025.1268
THINKMERGE Trimming (B)Model=Qwen3-4B, K=4, Strategy=All-Reduce2025.1268
GLM-4.5VModel Category=Large Vision-Language Models, Model Scale=108B2025.1267.34
THINKMERGE DirectMerge (A)Model=Qwen3-4B, K=2, Strategy=All-Reduce2025.1266.7
Majority Voting (Early-Ready)Model=Qwen3-4B, K=2, Strategy=Early-Ready2025.1266.7
THINKMERGE Early-Ready (C)Model=Qwen3-4B, K=2, Strategy=Early-Ready2025.1266.7
Qwen3-VL-32B-InstructModel Category=Large Vision-Language Models, Model Scale=32B, Reasoning Strategy=Instruct2025.1266.2
THINKMERGE Trimming (B)Model=Qwen3-4B, K=2, Strategy=All-Reduce2025.1266
THINKMERGE Shortest-K Merge (D)Model=Qwen3-4B, K=2, Strategy=Shortest-K Merge2025.1265.3
Majority VotingModel=Qwen3-4B, K=2, Strategy=All-Reduce2025.1263.3
JustRL-NemotronBackbone=OpenMath-Nemotron-1.5B, Sampling Strategy (@k)=@32, Dynamic Sampling=false, Training Steps=3440, Train Batch Size=256, Rollout N=8, Max Context Length=16k, Token Budget=1.1x10^8k2025.1262.92
QuestABackbone=OpenMath-Nemotron-1.5B, Sampling Strategy (@k)=@32, Dynamic Sampling=true, Training Steps=2000, Train Batch Size=128, Rollout N=16, Max Context Length=32k, Token Budget=2.6x10^8k2025.1262.08
Gemini 2.5 Flash2026.0260.3
Self-ConsistencyBase Model=Qwen3-30B-A3B, Zero-shot prompting=true, Thinking mode=false, Sampling=K=5 paths2026.0256.7
r1-distill-Llama-70Baccess=Open Weights, examples_for_finetuning=800K2025.0156.3
R1-distill-Llama-70B# Examples=800K2025.1256.3
s1K-1.1 + AIR-Step# Examples=1K, Distillation Source=Deepseek R12025.1253.33
Majority Voting (Shortest-K)Model=R1-Distill-Qwen-7B, K=8, Strategy=Shortest-K Merge2025.1252.7
THINKMERGE Shortest-K Merge (D)Model=R1-Distill-Qwen-7B, K=8, Strategy=Shortest-K Merge2025.1252.7
Majority Voting (Shortest-K)Model=R1-Distill-Qwen-7B, K=4, Strategy=Shortest-K Merge2025.1252
Skill-centric distillation frameworkBase Model=Qwen3-4B, SFT Setting=1K2026.0150.8
BaseBase Model=Qwen3-4B, SFT Setting=1K2026.0150.7
Majority Voting (Shortest-K)Model=R1-Distill-Qwen-7B, K=2, Strategy=Shortest-K Merge2025.1250.7
THINKMERGE Shortest-K Merge (D)Model=R1-Distill-Qwen-7B, K=4, Strategy=Shortest-K Merge2025.1250.7
RandomBase Model=Qwen3-4B, SFT Setting=1K2026.0150.3
s1Base Model=Qwen3-4B, SFT Setting=1K, Data Selection Mode=online2026.0150.2
s1.1access=Open Weights and Open Data, examples_for_finetuning=1K, budget_forcing=None2025.0150