model releases

OpenNavMap: Multi-Session Appearance-Based Topometric Mapping for Scalable Visual Navigation

arXiv:2601.12291v2 Announce Type: replace-cross Abstract: Scalable and maintainable maps are fundamental to large-scale navigation and the long-term deployment of robots in real-world environments. Ho

Model Releases

Operator-Informed Gaussian Processes for Complex Helmholtz Wavefields: From Synthetic Benchmarks to In Vivo Brain Elastography

DGX agent

arXiv:2607.14193v2 Announce Type: replace-cross Abstract: The Helmholtz equation governs time-harmonic wave propagation, and in dissipative media a complex modulus renders its squared wavenumber kappa

model-releasesarxiv-cs-lg

27 Jul 2026

Optimal Transport Image Representation and Deep Covariance Alignment (CORAL) for Control Valve Stiction Detection

arXiv:2607.22486v1 Announce Type: new Abstract: Control valve stiction is a common cause of unwanted oscillations and poor control-loop performance in industrial processes. Data-driven methods can aut

Optimization of time-consuming experimental conditions using pseudo-experimental data guided by adaptive polynomial regression

arXiv:2607.22238v1 Announce Type: new Abstract: Bayesian optimization (BO) is an optimization method that sequentially proposes the next candidate explainable variables for optimizing target variables

Parameter-free Adaptive Sparse Attention via Compression-Based Content Selection

arXiv:2607.21752v1 Announce Type: new Abstract: Data-adaptive sparse attention masks substantially outperform fixed patterns (e.g., BigBird and Longformer) and can even exceed dense attention on long

Pixels for Programs? A Cross-Provider Case Study of Input-Token Accounting for Source Code as Text and Images

arXiv:2607.21672v1 Announce Type: cross Abstract: Long source-code contexts consume many text tokens, motivating the proposal to render code as images for vision-language models. Recent work asks whet

PostDeg: Placement Beats Parameterization in LayerNorm GNNs

arXiv:2606.14022v2 Announce Type: replace Abstract: LayerNorm-based GNNs routinely erase the topology signals (degree, centrality, k-core) that node-selection policies should depend on, but the litera

Procedural Knowledge Is Not Low-Rank: Why LoRA Fails to Internalize Multi-Step Procedures

arXiv:2607.21612v1 Announce Type: cross Abstract: Parameter-efficient fine-tuning methods like LoRA have become the default for adapting large language models, succeeding across instruction following,

QC-PHAST Search: Classical--Quantum Query Benchmarks for Finite-Pool Rare-Regime Discovery

arXiv:2607.21995v1 Announce Type: cross Abstract: Rare-regime discovery in parameterized dynamical systems is an active-search problem: find one verified parameter at which a scientifically defined qu

Quasi-Monte Carlo Initialization for Meta-Reinforcement Learning

arXiv:2607.21637v1 Announce Type: new Abstract: This paper explores the efficacy of quasi-Monte Carlo (QMC) weight initialization for meta-reinforcement learning within modern benchmark environments.

Qwen3.6-27B speculative decoding gets better on heavier quants

I finished the speed leg of my spec-decode benchmarking for Qwen3.6-27B, main algorithms across quants. Overall: the heavier the quant, the more spec-decode buys you (10 of 10 speculative configs rank

RadSight: Towards Perceptually Reliable Multimodal Radiology Image Understanding

arXiv:2607.22293v1 Announce Type: new Abstract: Medical multimodal large language models (MLLMs) are increasingly expected to perform complex image understanding tasks, yet their reliability is often

ReCowGnition: A Realistic Biometric Benchmark for Cow Face Recognition

arXiv:2607.22071v1 Announce Type: new Abstract: With the development of precision livestock farming and the advances in computer vision, visual animal biometrics has gained attention. Using biometric

Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understa…

model-releasesclem-delangue--x

Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture:

Rethinking Layer-Wise Information Allocation for Vision Foundation Model Adaptation

arXiv:2607.21973v1 Announce Type: new Abstract: Vision foundation models are increasingly reused as frozen backbones for downstream visual recognition, making parameter-efficient adaptation a central

Risk-Routed Implicit Boundary Refinement for Robust Ultrasound Image Segmentation

arXiv:2607.21787v1 Announce Type: new Abstract: Medical ultrasound (US) image segmentation faces significant challenges due to speckle noise, low-contrast boundaries, acoustic shadowing, and acquisiti

Scaling Laws for Classical Machine Learning on Tabular Data: A Benchmark Study

arXiv:2607.21866v1 Announce Type: new Abstract: Prior classical-ML learning-curve work fits power laws to tree, linear, and kernel models on tabular data, but at small scale: typically one curve, one

SceneActBench: Can Agents Act on the 3D Scenes They See?

arXiv:2607.22393v1 Announce Type: cross Abstract: Vision-language model (VLM) agents increasingly use tools to act on 3D scenes rather than only describe them. Existing 3D benchmarks score textual res

SCOPE and SCION: A Benchmark and an Auditable Reference Pipeline for Schema Induction and Fusion from Text

arXiv:2607.21610v1 Announce Type: cross Abstract: Schema graphs are an upstream bottleneck of schema-grounded information extraction and knowledge graph construction, yet most extraction systems assum

Searching the Space of Feed-Forward Neural-Network Weight-Update Rules with Fixed Depth Symbolic Regression

arXiv:2607.21855v1 Announce Type: new Abstract: We investigate whether symbolic regression can discover explicit neural network weight-update rules that outperform standard hand-designed optimizers on

Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skills

arXiv:2607.22529v1 Announce Type: new Abstract: LLM training is shifting from manual design and annotation to interaction-driven self-evolution. However, existing self-evolutionary methods face a fund

Small context windows + knowledge graphs: the serialization format alone doubled my multi-hop accuracy (benchmarked 10 formats)

Running local models means every token counts — an 8K or 16K window fills up fast when you're stuffing graph context into prompts for RAG. I benchmarked 10 graph serialization formats (JSON, GraphML,

Spatially-Enhanced Temporal Fusion Transformer: Interpretable Multi-Output Prediction for Parametric Dynamical Systems with Time-Varying Inputs

arXiv:2505.00473v2 Announce Type: replace Abstract: We explore the promising performance of a transformer model in predicting outputs of parametric dynamical systems with external time-varying input s

Spectral Flow Certificates for Depth-Aware Long-Range Propagation in Graph Neural Networks

arXiv:2607.21607v1 Announce Type: cross Abstract: Graph Neural Networks propagate information through local message passing, but the graph topologies themselves can silently prevent any amount of trai

Stop to Decide: Latency-Aware Proprioceptive Navigation Primitives for Mapping-Free Quadruped Inspection

model-releasesarxiv-cs-ro

arXiv:2607.11204v2 Announce Type: replace Abstract: Onboard quadruped inspection systems often share limited compute between perception and navigation, reducing the rate at which event-triggered contr

SwiftMem: Fast Agentic Memory via Query-aware Indexing

arXiv:2601.08160v2 Announce Type: replace Abstract: Agentic memory systems have become critical for enabling LLM agents to maintain long-term context and retrieve relevant information efficiently. How

Synthetic Speech, Real Signal: Paralinguistic Preservation and Cross-Lingual Augmentation via Voice Cloning

arXiv:2607.22304v1 Announce Type: new Abstract: Synthetic data augmentation in speech is common practice for linguistic tasks like ASR, but has seen far less work for paralinguistic ones, especially c

The 3D Mirage: Probing and Taming 3D Hallucinations

arXiv:2512.15423v2 Announce Type: replace Abstract: Monocular depth foundation models achieve remarkable generalization by learning large-scale semantic priors, but this creates a critical vulnerabili

The Hard Decision Layer: Evidence for Committed Inference in Transformers

arXiv:2607.21613v1 Announce Type: cross Abstract: We investigate where and how transformer-based language models commit to predictions in multiple-choice question answering. We identify the _Hard Deci

The LangChain podcast where @hwchase17 interviews agent builders is full of alpha. Recent one with @EnoReyes was the best one yet. Sharing m…

model-releasesharrison-chase--x

The LangChain podcast where @hwchase17 interviews agent builders is full of alpha. Recent one with @EnoReyes was the best one yet. Sharing my unstructured notes: Eno keeps bringing back some core conc

The Lift Spectrum: How Measurement-to-Space Adaptivity Shapes Robustness in Image-Free Single-Pixel Sensing

arXiv:2607.22077v1 Announce Type: cross Abstract: Single-pixel sensing encodes a scene as a short sequence of coded measurements, and image-free methods infer the task directly from that sequence. Rem

Time-Reversed Imaging: A Multimodal Benchmark and Framework for Reconstructing Past Human-Environment Interactions

arXiv:2607.22352v1 Announce Type: new Abstract: We introduce time-reversed imaging, a new paradigm that infers what just happened in a scene from fading multimodal traces. Instead of extrapolating or

Toward Goal-Agnostic Joint-Embedding Predictive Control of Partial Differential Equations

arXiv:2607.21644v1 Announce Type: new Abstract: We present a goal-agnostic control framework for partial differential equations (PDEs) built around a joint-embedding predictive architecture (JEPA). Th

Toward User-Conditioned Evaluation of Personal LLM Agents under Temporal Interventions

arXiv:2607.21635v1 Announce Type: new Abstract: Personal agents maintain memories, learned skills, tool configurations, and policy state that evolve with each user. Existing agent benchmarks often eva

Trajectory-Regularized Stochastic Optimal Control via KL Divergence

arXiv:2607.22201v1 Announce Type: cross Abstract: We introduce trajectory-regularized stochastic optimal control (TRSOC), which augments standard stochastic optimal control (SOC) with a Kullback--Leib

Unbiased Open World Regularization for Fair Self-Supervised Learning

arXiv:2607.22149v1 Announce Type: new Abstract: Despite recent advances, self-supervised learning (SSL) models and Joint-Embedding Predictive Architectures (JEPAs) remain susceptible to learning spuri

Variational Low-rank Tensor Decomposition for Multisubject Spatiotemporal Data Analysis

arXiv:2607.22262v1 Announce Type: cross Abstract: Modeling shared and subject-specific structure in multisubject spatiotemporal data remains challenging, particularly in neuroimaging, where both spati

We are in a world where you can create truly unique, visually interesting and creative playable demos on demand with the current capabilitie…

model-releasesethan-mollick--x

We are in a world where you can create truly unique, visually interesting and creative playable demos on demand with the current capabilities of Codex and Claude Code. We don't need to keep cloning th

We are releasing PerceptionBench, a benchmark that isolates visual perception and evaluates it as a set of atomic capabilities - discovered …

model-releaseskimi-moonshot--x

We are releasing PerceptionBench, a benchmark that isolates visual perception and evaluates it as a set of atomic capabilities - discovered from how today's models fail, rather than defined in advance

We could really use Qwen3.8 in 27B, 35B, 122B and 397B sizes

Instead of 2T+ models, continuing to release highly capable small to medium size LLMs would really help to keep this community vibrant. Hardly anyone can even dream of running the recent 1.5-2T+ beast

We've open-sourced MoonEP, our high-performance communication library for distributed MoE workloads. Built to make expert-parallel communica…

model-releaseskimi-moonshot--x

We've open-sourced MoonEP, our high-performance communication library for distributed MoE workloads. Built to make expert-parallel communication more efficient at scale, MoonEP helps reduce communicat

WHBench: Evaluating Frontier LLMs with Expert-in-the-Loop Validation on Women's Health Topics

arXiv:2604.00024v2 Announce Type: replace Abstract: Large language models are increasingly used for medical guidance, but women's health remains under-evaluated in benchmark design. We present the Wom

You can now fine-tune my 3.96M-parameter TTS on your own voice or language

When I released Inflect v2 last week, I thought most people would ask whether a TTS model this small actually sounded decent. Instead, I kept getting two questions: “Can I train it on my own voice?” “

Zero-Shot Mission-Level Evaluation for Aerial MLLM Agents

arXiv:2607.22014v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) are emerging as core reasoning modules for embodied agents, yet it remains unclear how well general-purpose m

100B Models on Cheap Hardware: how realistic and limitations

There is a lot of buzz around running 100B parameter models on cheap local hardware using ternary (1.58-bit) quantization like Microsoft's BitNet architecture. The theoretical hardware shortcuts are i

model-releasesr-ollama

16 bit better than lower quants for Qwen3.6-27B

I am writing a fairly complex C++ windows MFC application. I have a few 3090s and can run F16 Qwen3.6-27B with 256K context and MTP. The quality of code is exceptional with this quant vs its lower qua

23 Gemma4-E4B models compared with abliterlitics: the most downloaded one is also the most broken

This is our biggest comparison yet. We've taken 23 Gemma 4 E4B models from huggingface and ran them through the abliterlitics gauntlet. We also have a new abliterlitics discord, feel free to jump on a

90 agentic bakeoff runs: ThinkingCap vs Fable Fusion vs stock Qwen3.6-27B

Last week someone here said ThinkingCap and Fable Fusion 'really do beat the OG' for agentic work, so I ran it: 6 self-grading tasks, 5 reps, 3 models, 90 isolated runs. Tooling, since that's half the