DailyPapers
@HuggingPapers
Tweeting interesting papers submitted at
Submit your own at and link models/datasets/demos to it!
4 Following 22.2K Followers
RoTA-Bench: a new benchmark for 3D spatial reasoning
It challenges models to align projections by rotating objects in 3D from 2D views. Humans solve ~100% of tasks; GPT-5.6-Sol only ~17%.
Show more
NVIDIA just released a LiLiCorr speculative decoding draft for Nemotron 3.5 Bolt on Hugging Face
It pairs with the 74B target to accept longer token blocks and serve faster, while the target verifies every token.
Show more
Doc2Harness: The first benchmark for long-horizon harness engineering of coding agents
19 frontier models, 328 test cases, best pass rate only 53% — even the strongest fails half. From ICT-CAS.
Show more
Mistral just released Timestral, a time series foundation model, on Hugging Face
State-of-the-art zero-shot forecasting across 3 major benchmarks, with 21 quantiles and up to 32k context.
Show more
NVIDIA just released a GR00T-based robotics model on Hugging Face
Agile One S SSD Pick is a deployment model for SSD picking, with ONNX graphs and TensorRT engines included.
NVIDIA just released a GR00T N1.7 robotics deployment model for SSD pickup on Hugging Face
NVIDIA just released Long-WAM
A world-action model that scales visual context for real-time robot control. Access to history is not the same as using it: longer histories pay off far more when the video foundation is pretrained autoregressively.
Show more
Tencent researchers just released STEPQuant on Hugging Face
A spatial-temporal post-training quantization framework for Delta-rule recurrent states, matching FP32 accuracy with 6-bit states while cutting total serving memory by up to 68.7%
Show more
ID-Forcing: keeping long video generation in-distribution
Autoregressive video models drift beyond their training horizon. ID-Forcing aligns KV caching and conditioning with training, enabling minute-scale videos with no fine-tuning.
Show more
Fold2Reason: training on protein folding improves broad reasoning
Post-training on FoldingCorpus raises macro-average accuracy from 45.09% to 48.33% across 10 reasoning benchmarks, with 2.7–3.5× better structure prediction.
Show more
AREX-2: Advancing Self-Improving Agents through Long-Horizon Reflective Tasks
A 27B agent that turns more test-time rounds into better solutions, trained on verifiable ML and algorithmic tasks. Achieves 81.8 on MLE-bench Lite, 70.7 on Frontier-CS, and transfers to deep research.
Show more
The Teacher Is a Direction, Not a Destination
New on-policy distillation method that extrapolates RL-induced representation residuals, letting students match or exceed the teacher across four model pairs. Code and checkpoints planned for release.
Show more
Post-Training Leaves Behavioral Shadows on Unrelated Decisions
With 5,664 single-word teacher answers, a student gains +5.34pp on HumanEval+ — without ever seeing code.
InternW0-Δ: a world action model for robots
Pretrained on over 20,000 hours of open robot and human data, this model jointly learns predictive dynamics and action generation, outperforming prior methods across benchmarks and real robots.
Show more
Alibaba releases a new training framework for software engineering agents
It develops category experts via iterative Refresh–Repair–Expand (RRE) and consolidates them into one deployable model with label-routed multi-teacher distillation (MOPD). Achieves 58.04% on Pro-618 and 59.00% on SWE-bench Multilingual.
Show more
SpeakerMem-R1
A speaker-centered dual-track memory for multi-party dialogue that stores verbatim messages and organizes derived states by person and group, tackling attribution and state reconstruction.
Show more
Microsoft just released rho-roboeval on Hugging Face
A new model for robotics evaluation.
RoboDawn: Transferring the Intelligence of VLMs to Robotic Control
Tsinghua and Tencent Hunyuan show a frozen VLM can drive robots through simple move, rotate and gripper commands, reaching 73.6% one-shot success on RoboTwin 2.0 C2R.
Show more
Tencent ARC Lab just released GameHorizon Suite
A unified data and evaluation suite measuring AAA gameplay capabilities across multiple temporal horizons for VLMs, UMMs, GUI, coding, and game agents.
Show more
Alibaba's Qwen team just released RecreationWorld on Hugging Face
A five-platform sandbox where hybrid computer-use agents explore real apps, implement them, and visually verify their own builds.
Show more
