ICLR Outstanding Paper Awards winners by year

ICLR · Paper award · usually Apr or May · next expected Apr 2027

Winners

Transformers are Inherently SuccinctOutstanding paperSource ↗

LLMs Get Lost In Multi-Turn ConversationOutstanding paperSource ↗

The Polar Express: Optimal Matrix Sign Methods and their Application to the Muon AlgorithmHonorable mentionSource ↗

Safety Alignment Should be Made More Than Just a Few Tokens DeepOutstanding paperSource ↗

Learning Dynamics of LLM FinetuningOutstanding paperSource ↗

AlphaEdit: Null-Space Constrained Model Editing for Language ModelsOutstanding paperSource ↗

Data Shapley in One Training RunHonorable mentionSource ↗

SAM 2: Segment Anything in Images and VideosHonorable mentionSource ↗

Faster Cascades via Speculative DecodingHonorable mentionSource ↗

Generalization in diffusion models arises from geometry-adaptive harmonic representationsOutstanding paperSource ↗

Learning Interactive Real-World SimulatorsOutstanding paperSource ↗

Never Train from Scratch: Fair Comparison of Long-Sequence Models Requires Data-Driven PriorsOutstanding paperSource ↗

Protein Discovery with Discrete Walk-Jump SamplingOutstanding paperSource ↗

Vision Transformers Need RegistersOutstanding paperSource ↗

Amortizing intractable inference in large language modelsHonorable mentionSource ↗

Approximating Nash Equilibria in Normal-Form Games via Stochastic OptimizationHonorable mentionSource ↗

Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN ExpressivenessHonorable mentionSource ↗

Flow Matching on General GeometriesHonorable mentionSource ↗

Is ImageNet worth 1 video? Learning strong image encoders from 1 long unlabelled videoHonorable mentionSource ↗

Meta Continual Learning Revisited: Implicitly Enhancing Online Hessian Approximation via Variance ReductionHonorable mentionSource ↗

Model Tells You What to Discard: Adaptive KV Cache Compression for LLMsHonorable mentionSource ↗

Proving Test Set Contamination in Black-Box Language ModelsHonorable mentionSource ↗

Robust agents learn causal world modelsHonorable mentionSource ↗

The mechanistic basis of data dependence and abrupt learning in an in-context classification taskHonorable mentionSource ↗

Towards a statistical theory of data selection under weak supervisionHonorable mentionSource ↗

About this award

Outstanding papers and honorable mentions at the International Conference on Learning Representations.

Edition	Announced	Winners
2026	Apr 23, 20266 months ago	3
2025	Apr 22, 20251 year ago	6
2024	May 6, 20242 years ago	16

The next date is an estimate from the months past editions were announced (Apr or May); it changes when ICLR announces one.

Date announcedSep 25, 2026Next edition expected Apr 2027 (was Apr 2026).

New winnersSep 25, 2026Editions 2024 to 2026 loaded with 25 winners; the 2026 edition named Transformers are Inherently Succinct and 2 others.

Award addedSep 25, 2026Now tracked: Outstanding papers and honorable mentions at the International Conference on Learning Representations.