AI models and papers trending on September 29, 2026
The edition of this day as it was recorded: the most upvoted papers of the daily list.
On Tuesday, September 29, 2026 the most upvoted paper was "Post-Training Leaves Behavioral Shadows on Unrelated Decisions" from Peking University, with 271 votes. 10 of the 15 leading papers had a code repository attached. No model list exists for this day: Hugging Face does not publish past trending lists, and our collector had not started yet.
Papers only · rebuilt on 2026-10-03 09:16 UTC from the Hugging Face daily papers archive
Papers
- 271
Post-Training Leaves Behavioral Shadows on Unrelated Decisions
Peking UniversityarXiv 2609.29233 (opens in a new tab)Code (opens in a new tab)
- 241
YuE2: Unifying Symbolic and Audio Music Generation at Frontier Quality
Multimodal Art ProjectionarXiv 2609.33757 (opens in a new tab)Code (opens in a new tab)
- 215
VisionHOPE: Visual Backbones as Self-Modifying Learning Systems
Mininglamp TechnologyarXiv 2609.33325 (opens in a new tab)Code (opens in a new tab)
- 176
Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation
Nanyang Technological UniversityarXiv 2609.35347 (opens in a new tab)Code (opens in a new tab)
- 128
Groupwise Agentic Grading and Advantage Redistribution for Code Agent RL
Xiaomi MiMoarXiv 2609.32577 (opens in a new tab)
- 106
Self-Evolving Coding Agents: From Digital Programs to Physical-World Intelligence
HexaFuturearXiv 2609.35432 (opens in a new tab)Code (opens in a new tab)
- 89
Duplex-MPE: Benchmarking Multi-Party Interaction in Full-Duplex Dialogue
Peking UniversityarXiv 2609.31948 (opens in a new tab)
- 75
MassAlloc Attention: Let Attention Allocate Its Own Compute
Beijing Academy of Artificial IntelligencearXiv 2609.32712 (opens in a new tab)Code (opens in a new tab)
- 70
TraceDance: An Automated System for Building Agent Behavior Benchmarks from Real-World Agent Deployment Traces
ByteDancearXiv 2609.33295 (opens in a new tab)Code (opens in a new tab)
- 66
CoWindow Attention: Full Causal Coverage Is a Collective Property
Beijing Academy of Artificial IntelligencearXiv 2609.32704 (opens in a new tab)
- 65
How Far Are We from Removing the Visual Encoder? Scaling Laws for Encoder-Free Multimodal Pretraining
Tencent HunyuanarXiv 2609.35457 (opens in a new tab)
- 55
Improving Test-Time Scaling with Adaptive Looped Transformers
Tsinghua-NICS-EFCarXiv 2609.35748 (opens in a new tab)Code (opens in a new tab)
- 48
Learning Native Reflection in Unified Models with Interleaved Reinforcement Learning
MMLab@NTUarXiv 2609.35767 (opens in a new tab)Code (opens in a new tab)
- 42
Surprising Success, Repeated Failure: Entropy-Guided Credit Assignment for Exploration in LLM Reasoning
KAIST AIarXiv 2609.33781 (opens in a new tab)Code (opens in a new tab)
- 41
QwenGyre: An Elastic Reinforcement Learning Framework for Training xLong-Horizon Agents
Models
No model list exists for this day. Hugging Face does not publish past trending lists, so days before the collector started carry papers only.
About the archive
How far back does the archive go?
Model lists exist from the first day the collector ran: Hugging Face does not publish past trending lists, so they cannot be rebuilt. Paper lists for earlier days were rebuilt from the Hugging Face daily papers archive, and those days are marked "papers only".
What is kept for each day?
The leading models with their task, likes and downloads on that day, the models held back, and the leading papers with their upvotes. Abstracts are not duplicated in the archive; each paper links to arXiv.
Do archived numbers change?
No. A day’s record is the last collection of that day and is not rewritten afterwards. Likes and downloads you see on an archived day are the values on that day, not today’s.
How are weeks defined?
ISO 8601 weeks: Monday to Sunday, numbered so that week 1 contains the first Thursday of the year. The address /week/2026-40 is week 40 of 2026.