Project AURIONAI Use-Cases Research & Innovation

AIME 2025

AIME 2025 problems

2025mathsaturated

Score in %

  1. 1Technology Innovation Institute (TII)Falcon-H1R-7B83.1 %vendor-reported
  2. 2Allen AIOlmo 3 32B Think72.5 %vendor-reported
  3. 3Hugging FaceSmolLM3-3B36.7 %vendor-reported
13scores
12models
0independent
0maintainer-run
13vendor-reported

Definition

Version
2025
Category
math
Maintainer
not in the notes
Status
saturated
Home page
not in the notes

Sources

  1. allenai/Olmo-3-32B-Think , Hugging Face huggingface.co https://huggingface.co/allenai/Olmo-3-32B-Think
  2. huggingface.co/microsoft/Phi-4-mini-flash-reasoning https://huggingface.co/microsoft/Phi-4-mini-flash-reasoning
  3. HuggingFaceTB/SmolLM3-3B , Hugging Face huggingface.co https://huggingface.co/HuggingFaceTB/SmolLM3-3B

Scores at a Glance

Score in %

Higher is better for this unit.

1Technology Innovation Institute (TII)Falcon-H1R-7BTechnology Innovation Institute (TII)
83.1 %
vendor-reported2Allen AIOlmo 3 32B ThinkAllen AI
72.5 %
vendor-reported3Hugging FaceSmolLM3-3BHugging FaceCard 3-model comparison table; mode label not extracted
36.7 %
vendor-reported4Microsoft AIPhi-4-mini-flash-reasoningMicrosoft AI
33.59 %
vendor-reported
0255075100
Where this comes from
Each bar is one model, with its most independent recorded score. Hatched bars are vendor-reported; the word beside each bar says who ran it. Settings and dates differ between rows; the Scores table below has every result.

Scores

13 of 13 scores shown aime-2025-scores.csv
ModelScoreWho ran itSettingView or populationDateSource
Arcee AITrinity-Large-Thinking96.3 %vendor-reported–Trinity-Large-Thinking model card1 Apr 2026huggingface.co
XiaomiMiMo-V2-Flash94.1 %vendor-reported–MiMo-V2-Flash model card, post-trained model table16 Dec 2025huggingface.co
DeepSeekDeepSeek-V3.188.4 %vendor-reportedThinking mode, Pass@1DeepSeek-V3.1 model card21 Aug 2025huggingface.co
LG AI ResearchEXAONE 4.0 (32B)85.3 %vendor-reportedReasoning modeEXAONE 4.0 model cardJul 2025huggingface.co
Technology Innovation Institute (TII)Falcon-H1R-7B83.1 %vendor-reported–––huggingface.co
Allen AIOlmo 3 32B Think72.5 %vendor-reported–––huggingface.co
NVIDIALlama 3.1 Nemotron Ultra 253B v172.5 %vendor-reportedReasoning on, pass@1Nemotron Ultra model cardApr 2025huggingface.co
inclusionAILing-1T70.42 %vendor-reportedAverage over 4,000 output tokens per problemAnt Group claim reported by AI NewsOct 2025artificialintelligence-news.com
NVIDIALlama 3.3 Nemotron Super 49B v158.4 %vendor-reportedReasoning on, pass@1Nemotron Super model cardMar 2025huggingface.co
Allen AIOlmo 3.1 32B Instruct57.9 %vendor-reported–Olmo 3.1 32B Instruct model cardDec 2025huggingface.co
Hugging FaceSmolLM3-3B36.7 %vendor-reportedCard 3-model comparison table; mode label not extracted––huggingface.co
Microsoft AIPhi-4-mini-flash-reasoning33.59 %vendor-reported–––huggingface.co
Hugging FaceSmolLM3-3B9.3 %vendor-reportedCard 5-model comparison table; mode label not extracted––huggingface.co
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "AIME 2025", https://aurion.semai.id/benchmarks/aime-2025/, records of October 2026.