Eight universities share the lead with two major AI paper awards since 2024

Across verified Best Paper and Outstanding Paper winners at flagship ML, NLP and vision conferences, no university appears on more than two winning papers. Princeton, MIT, NUS, Oxford, Edinburgh, Washington, Tel Aviv and NYU form the leading group under full-count credit for any paper with at least one primarily affiliated author.

2
winning papers per leader
8
universities tied for first
16
university–paper links
6
flagship venues represented
Award-winning papers credited to each leading university, 2024–25
papers
Princeton
2
MIT
2
NUS
2
Oxford
2
Edinburgh
2
Washington
2
Tel Aviv
2
NYU
2

What drives the tie

01
Princeton spans safety and deep reinforcement learning. Its ICLR 2025 Outstanding Paper and NeurIPS 2025 Best Paper provide two independent wins. proceedings.iclr.cc
02
MIT and Edinburgh each bridge two research communities. MIT links world models with language-model attention; Edinburgh links cross-lingual NLP with attention architecture. arxiv.org
03
NUS wins in consecutive years. The university appears on NeurIPS 2024’s stochastic-derivative winner and ICLR 2025’s model-editing winner. neurips.cc
04
Full credit makes collaboration visible. Each affiliated university receives one count per paper; corporate and independent-lab affiliations are shown but do not enter the university ranking. github.com

Winning papers and primary affiliations

UniversityPaperVenueAwardPrimary author affiliations
PrincetonSafety Alignment Should Be Made More Than Just a Few Tokens DeepICLR 2025Outstanding PaperPrinceton University; Google DeepMind
Princeton1000 Layer Networks for Self-Supervised RLNeurIPS 2025Best PaperPrinceton University; Warsaw University of Technology; Tooploox; IDEAS Research Institute
MITLearning Interactive Real-World SimulatorsICLR 2024Outstanding PaperUC Berkeley; Google DeepMind; MIT; University of Alberta
MITGated Attention for Large Language ModelsNeurIPS 2025Best PaperAlibaba/Qwen; University of Edinburgh; Stanford; MIT; Tsinghua
NUSStochastic Taylor Derivative EstimatorNeurIPS 2024Best PaperNational University of Singapore; Sea AI Lab
NUSAlphaEdit: Null-Space Constrained Model EditingICLR 2025Outstanding PaperNational University of Singapore; University of Science and Technology of China
OxfordThe PRISM Alignment DatasetNeurIPS 2024Best PaperUniversity of Oxford and collaborating research organizations
OxfordVGGT: Visual Geometry Grounded TransformerCVPR 2025Best PaperUniversity of Oxford; Meta AI
EdinburghBridging the Language Gaps in Large Language ModelsACL 2025Outstanding PaperUniversity of Edinburgh
EdinburghGated Attention for Large Language ModelsNeurIPS 2025Best PaperAlibaba/Qwen; University of Edinburgh; Stanford; MIT; Tsinghua
WashingtonArtificial HivemindNeurIPS 2025Best PaperUniversity of Washington; Allen Institute for AI; collaborating universities
WashingtonNative Sparse AttentionACL 2025Best PaperPeking University; DeepSeek-AI; University of Washington
Tel AvivNever Train from ScratchICLR 2024Outstanding PaperTel Aviv University; IBM Research
Tel AvivInformation Complexity of Stochastic Convex OptimizationICML 2024Best PaperTel Aviv University; University of Toronto; Google Research
NYUGeneralization in Diffusion Models Arises from Geometry-Adaptive Harmonic RepresentationsICLR 2024Outstanding PaperNew York University; Flatiron Institute; ENS/PSL
NYUProtein Discovery with Discrete Walk-Jump SamplingICLR 2024Outstanding PaperGenerate Biomedicines; New York University; collaborating institutions

Method: live-web review run 25 Aug 2026 over 1,839 unique search results and 97 extracted award mentions, then deduplicated by paper. Scope is conference-level Best Paper and Outstanding Paper at ICLR, ICML, NeurIPS, ACL, CVPR and ECCV from 2024 onward; honorable mentions, student, workshop, dataset-only and test-of-time prizes are excluded. A paper gives one full count to every university appearing as a primary author affiliation, so totals are university–paper links rather than fractional credit. The 2026 cycle is incomplete and affiliations are normalized to parent universities.

Keenable · made with SELECT* · 7,179 pages in 5m 18s · Open the chatShare: XLinkedInReddit