Model releases drive changes across AI products and infrastructure. This hub tracks frontier models, multimodal capabilities, open weights, context windows, benchmark signals, API changes, and deployment paths so readers can judge whether a new model changes cost, quality, or availability.
Zyphra released ZUNA1.1 on July 16, 2026, under the Apache 2.0 license. The 380M masked diffusion autoencoder reconstructs, denoises, and upsamples scalp-EEG across arbitrary channel layouts. It accepts variable-length inputs from 0.5 to 30 seconds, against ZUNA1's fixed five seconds. Reported NMSE holds or improves while the input range widens.
ZUNA1.1 accepts variable-length inputs from 0.5 to 30 seconds, tokenized into 0.125-second segments.
It uses a transformer encoder-decoder with 4D RoPE and rectified flow objective.
In a recent benchmark, GPT-5.6 Sol Ultra autonomously constructed a complete Chrome V8 exploit chain from scratch by analyzing security-fix patches, ultimately popping a calculator. Other frontier models like Sol Medium and Grok 4.5 stalled early. The author argues that exploit development as a human skill is now obsolete.
GPT-5.6 Sol Ultra completed a 9-step exploit chain in three days, including Maglev type confusion, sandbox read/write, sandbox escape, UAF, and code execution.
Sol Medium and Grok 4.5 failed to advance beyond sandbox primitives; Sol Ultra used 74 sub-agents and 2.1B tokens at a cost of ~$1,597.
The chipmaker is fleshing out its physical AI ecosystem, from foundation models and edge hardware to software, developer tools and industrial partnerships.
Nvidia expands physical AI strategy covering robotics and edge computing
Introduces foundation models and edge hardware for AI applications
Meta's new Muse Spark 1.1 model is now available on Databricks via Model Provider Services (MPS) in Unity AI Gateway. This service allows organizations to register providers once in Unity Catalog, eliminating API key sprawl and centralizing governance through familiar permissions, rate limits, and guardrails. Additionally, every request is automatically tracked with token usage, latency, cost attribution, and audit logs for end-to-end observability.
Access Meta's new Muse Spark 1.1 model on Databricks through Model Provider Services in Unity AI Gateway.
Register providers once in Unity Catalog to centralize access, rate limits, and guardrails.
Simon Willison developed a tool to detect and highlight common clichéd phrases often found in LLM-generated text, such as 'no fluff, no filler, no jargon'. The tool runs in the browser, provides pattern counts and navigation, and aims to reduce frustration with formulaic AI writing.
Simon Willison created the LLM cliché highlighter to identify overused phrases in AI-generated content.
The tool highlights patterns like 'no X, no Y' chains and 'you already know'.
OpenTSLM is a multimodal LLM that treats time series as a native modality, enabling reasoning over raw multivariate signals alongside text. It outperforms baselines, including GPT-4o, on time series QA, activity recognition, sleep staging, and ECG QA. The model scales to multiple long time series with near-constant memory consumption. ECG reasoning validated by 7 cardiologists with 97% correctness. All code, datasets, and models are open-source.
OpenTSLM is a multimodal LLM that natively processes time series alongside text for reasoning.
It surpasses GPT-4o and other baselines on several time series tasks, even at 1B parameters.
NVIDIA released Nemotron 3 Embed on July 15 and 16, 2026. The collection has three open checkpoints: Nemotron-3-Embed-8B-BF16, Nemotron-3-Embed-1B-BF16, and Nemotron-3-Embed-1B-NVFP4. The 8B ranks #1 on RTEB at 78.46 average NDCG@10. The 1B came from ModelOpt NAS pruning plus COS+MSE distillation from the 8B teacher. NVFP4 retains 99%+ of BF16 retrieval accuracy at up to 2x Blackwell throughput. All three run 32,768-token inputs under OpenMDW-1.1.
Nemotron-3-Embed-8B-BF16 ranks #1 on RTEB with 78.46 average NDCG@10
Three open checkpoints: 8B BF16, 1B BF16, and 1B NVFP4
This paper proposes ConFlow, a framework that incorporates constraint information directly into the flow matching training objective via differentiable barrier or cost functions and a conditional Gaussian Process, improving constraint satisfaction and trajectory quality in robot motion generation. Experiments on a two-robot navigation task demonstrate lower collision rates and higher trajectory quality compared to standard flow matching baselines.
ConFlow bridges the training-inference gap by integrating differentiable constraint functions into the training objective
Replaces standard Gaussian source distribution with a conditional Gaussian Process to handle smoothness and boundary conditions
This paper explores the feasibility of using brain signals via functional near-infrared spectroscopy (fNIRS) to modulate robot reinforcement learning. It compares agents trained on passive (observational) versus active (demonstrative) interaction tasks, and tests multiple methods for enhancing the RL algorithm with the neural signal, focusing on parameter augmentation rather than replacement. The results show that this framework is effective: the neural signal improves learning when augmenting trajectory priorities and state-action q-values. Additionally, the framework learns successfully from offline data, offering a practical alternative for settings where real-time BCI setups are impractical or only limited data is available.
fNIRS brain signals can enhance robot reinforcement learning
Comparison of passive and active interaction tasks
DiMaS is a distribution-matching steering strategy for flow-matching vision-language-action (VLA) models, enabling fine-grained behavioral control in robotic manipulation. It transports between representation distributions rather than shifting along a fixed direction, proving effective on two state-of-the-art VLAs. The study also examines transferability and explains why linear steering fails in visuomotor settings: behavioral features are linearly decodable but not linearly steerable.
DiMaS achieves fine-grained behavioral control by transporting between representation distributions instead of linear shifts.
It works on two SOTA VLAs, with analysis of how task similarity affects control transfer.
MEMORA introduces Embodied Action Memory (EAM) to enable robots to use persistent memory from egocentric video for long-horizon planning. It features four typed memory stores, online editing, and offline consolidation. Evaluated on 45 hours of EPIC-KITCHENS-100 video, MEMORA improves memory accuracy by up to 20.5 points and planning scores by 16.6%.
Embodied Action Memory (EAM) for long-horizon robot planning.
Four memory stores: Environment, Entity, Activity, Inferred Knowledge.
This paper proposes LIFT, a force-aware post-training framework that adds contact reactivity to pretrained vision-language-action (VLA) policies. By grafting a reactive action expert, injecting 6D end-effector force via causal force memory and cross attention, and coupling with an online DAgger loop, LIFT outperforms vision-only post-training in towel folding, book insertion, and Hanoi ring placement.
LIFT enhances VLA policies with contact reactivity while preserving general manipulation knowledge.
It uses a reactive action expert, causal force memory, and online DAgger training to handle distribution shifts.
This work introduces a multi-modal orchestration framework for semantic audio-driven humanoid control, enabling real-time autonomous selection of motion skills based on music or speech input. Validated on the Unitree G1 humanoid, it demonstrates robust sim-to-real transfer.
Proposes a semantic audio-driven framework for humanoid whole body control with real-time skill selection.
Processes music via audio fingerprinting and speech via imitation-learned skill library.
SD-MAR is a framework for training and evaluating vision-language models (VLMs) on multi-image analytical reasoning tasks. It constructs paired visual scenarios through controlled perturbations and generates reasoning tasks spanning semantic change attribution and quantitative comparison. Using GRPO-lite with Backward Discounted Allocation (BDA), a reinforcement learning approach that removes KL regularization, fine-tuning on SD-MAR improves in-domain accuracy by up to 36.95% on Qwen2.5-VL-7B and InternVL3-8B. Qwen2.5-VL-7B outperforms GPT-4.1 on the SD-MAR benchmark. Out-of-domain generalization is preserved or improved, with performance within 1% on MME, MMMU-Pro, MathVista and up to 4% improvement on MMBench. LLM-as-judge evaluation shows consistent improvements in logical coherence and explanation quality.
SD-MAR generates multi-image reasoning tasks via synthetic data.
GRPO-lite with BDA reinforcement learning enhances policy optimization.
Addressing the challenge of defect segmentation in additive manufacturing XCT images, the proposed XCT-SAM framework sequentially adapts SAM using Conv-LoRA adapters, first on an alloy microstructure dataset then on XCT images, outperforming baselines on CycleGAN-XCT benchmarks and real NIST scans.
XCT-SAM performs two-stage domain adaptation, fine-tuning Conv-LoRA on alloy microstructure data before transferring to XCT images.
Only about 4.15 million parameters are trained, with over 99% of the model frozen.
MonteRET is a region-aware retrieval-enhanced framework for generating chest CT findings sections. It integrates global and regional CT features, retrieves clinically relevant knowledge, and refines reports via a knowledge-guided rewriting agent. Evaluated on public and external cohorts, MonteRET improved report quality, semantic similarity, and clinical efficacy, with experts favoring its outputs.
MonteRET combines global CT features with region-level representations and retrieves knowledge using predicted conditions and vision-language alignment.
Trained on 24,128 CT scans and evaluated on 1,564 public test scans plus 82 external scans.
This paper investigates whether vision foundation models build representations that reflect intrinsic properties of 3D Euclidean space. Instead of regressing depth or normals, the authors probe the relationship between visual feature space structure and Euclidean transformation group SE(3) using a mutual neighborhood metric and a Poincaré Adapter. They show that self-supervised vision models harbor latent subspaces strongly correlated with 3D space, even without 3D supervision. This leads to 'Latent-Space Navigation' techniques for visual odometry and localization without explicit 3D reconstruction.
Probes the 3D awareness of vision features from topological and geometric perspectives
Introduces mutual neighborhood metric and Poincaré Adapter as evaluation tools
The paper presents KeyFrame-Compass, the first comprehensive benchmark for evaluating keyframe-conditioned video generation, with 386 curated samples, an automated evaluation framework, and experiments revealing trade-offs between keyframe fidelity and naturalness.
KeyFrame-Compass is the first comprehensive benchmark for keyframe-conditioned video generation.
It includes 386 samples across diverse settings and an automated evaluation framework with six metrics for keyframe execution.
Multi-reference-to-audio-video (MR2AV) generation requires models to produce synchronized audio-video content conditioned on multiple references and textual instructions. Existing benchmarks focus on text-driven generation or single-reference preservation, lacking evaluation for MR2AV. This paper introduces MultiRef-Compass, a unified benchmark with 350 carefully curated samples covering multi-view subject preservation, multi-entity binding, and human-object-scene composition. It defines an evaluation protocol with four dimensions (Basic Quality, Reference Consistency, Audio-Visual Consistency, Instruction Following) and 14 sub-metrics, integrating automatic metrics with a rejudging-enhanced MLLM-as-a-Judge framework. Experiments on eight MR2AV systems reveal substantial room for improvement across all dimensions.
MultiRef-Compass is the first comprehensive benchmark for MR2AV generation, comprising 350 samples.
It covers multi-view subject preservation, multi-entity binding, and human-object-scene composition, with a four-dimensional evaluation protocol (14 sub-metrics).
This paper introduces tool efficiency, a new quantitative metric to evaluate the rate of useful tool calls in an LLM agent trajectory. To ensure that tool efficiency is well-defined, it also introduces marginal tool utility, indicating per tool call whether it is useful or safely removable. The sign of marginal tool utility is determined using LLM-as-a-Judge. This work directly measures efficiency, complementing accuracy-based evaluations, and aims to inform future benchmark design and lean tool suite engineering.
Introduces tool efficiency as a metric for useful tool call ratio in LLM agent trajectories.
Defines marginal tool utility to assess necessity of individual tool calls.
Polestar is a training-free inference framework that addresses KV-cache reuse and decoding parallelism challenges in diffusion LLMs by leveraging token representation drift. It consists of Polestar-Cache for sparse cache refreshes and Polestar-Commit for identifying commit-ready tokens, achieving up to 10.73% accuracy improvement and 3.7x higher throughput on math and coding benchmarks.
Polestar uses token representation drift to jointly optimize cache efficiency and decoding parallelism.
Polestar-Cache identifies stale KV-cache positions for sparse refreshes, enabling efficient reuse.
This paper introduces token time continuous diffusion (TTCD), a diffusion language model operating in continuous space with per-token times, where tokens proceed from noise to token at varying rates. TTCD avoids parallel sampling inaccuracies and outperforms discrete models at high speedups. A 160M parameter model trained on OpenWebText and self-distilled achieves comparable unconditional and superior conditional generation, with gains in Sudoku solving.
TTCD is a continuous-space diffusion LM with per-token times, allowing tokens to be generated at different rates.
Continuous space avoids inaccuracies from parallel sampling, improving performance at high speedups.
The paper introduces AGOPS, an automatic method to generate task-specific prompt guidelines that help users write better prompts, improving LLM performance by recovering large performance drops from underspecification.
Underspecified prompts cause up to 95.3% performance drop in LLMs.
Existing prompt guidelines are generic and manually created.
A new study quantifies information loss when LLM agents communicate via text, using sparse autoencoder feature analysis. While latent communication preserves more information under compression, the lost features primarily encode surface form rather than task-relevant semantics, questioning the practical advantage of latent channels.
SAE-sparse channel retains 99.4% probe accuracy at 28x compression vs 80.4% for text.
Cross-architecture latent alignment achieves 92% top-1 retrieval between Llama and Mistral.
This paper proposes LBA, a sampling-based method for generating high-quality adversarial texts under low query budgets in the hard-label setting. By integrating prior and posterior knowledge to construct an approximate distribution, LBA efficiently samples adversarial examples. Extensive experiments show LBA outperforms state-of-the-art baselines across models and datasets, with better semantic preservation and readability.
Existing hard-label attacks use greedy algorithms, leading to high query costs and suboptimal solutions.
LBA uses sampling with an approximate distribution updated by posterior knowledge.
This paper presents the first application of pregroup grammar-based quantum compositional NLP to Arabic, a morphologically rich language. Quantum circuits mirror grammatical structure, outperforming classical baselines in word order, tense, and verb sense disambiguation experiments.
First QNLP application to Arabic using pregroup grammar.
Sentences converted to quantum circuits reflecting grammatical topology.
The Just Keep Prompting (JKP) framework tests VLM stability under repeated challenging. Evaluations on GPT-4o, Gemini 2.5 Pro, and Qwen3-VL-30B show substantial instability and answer flipping, with model-specific pressure-response profiles.
JKP uses three strategies (Adversarial Negation, Pure Socratic Interrogation, Context-Aware Socratic Summarization) to probe models over up to 10 turns.
Aggregate accuracy changes little, but trajectory analysis reveals frequent answer flips and instability.
This paper analyzes why closed-loop knowledge systems (e.g., LLMs, RL) saturate under repeated internal feedback and introduces a three-level operational framework to enable escape via structural interventions. Using Lyapunov drift, stability is characterized, and escape is quantified by attractor displacement and a KL lower bound. Case studies include LLM code repair, sparse-reward RL, and Bayesian optimization.
Closed-loop systems exhibit diminishing returns under repeated internal feedback; external information is needed to escape attractors.
A three-level framework is proposed: knowledge states evolve via transition kernels indexed by structural parameter θ; interventions change θ and are falsifiable.
Offline reinforcement learning world models suffer from model exploitation in low-data regions. RENEW uses human preferences over imagined rollouts to directly repair exploitation, introducing epistemic uncertainty to focus finetuning and improve sample efficiency.
World models in offline RL are vulnerable to exploitation in thin data coverage regions.
RENEW leverages human preferences to identify and fix dynamics hallucinations.
Proposes Branching Policy Optimization (BPO), which leverages deterministic, snapshottable, and resumable sandboxes to construct a tree-structured rollout topology with shared prefixes, reducing policy gradient variance and improving success rates by 3.6–6.1 absolute points over GRPO and RLOO.
BPO exploits sandbox properties to create a tree of trajectories with shared prefixes, replacing independent trajectory sampling.
It branches at decision points and computes advantages from sibling returns, provably reducing variance compared to trajectory-level baselines.
This paper introduces C3R, a drop-in control layer that, from an inferred domain posterior and no query-time label, certifies a per-domain contamination budget where feasible and otherwise abstains. It guarantees a reduction on the hardest domains, shows stability across resampling, and retains more recall than calibrated cascades.
C3R provides label-free per-domain contamination control with conformal risk guarantees.
It uses a two-split scheme with finite-sample transfer bounds that support heterogeneous budgets.
Existing zero-shot image classification methods using vision-language models (VLMs) often employ a uniform weighting of prompts across all classes, ignoring the class-specific suitability of prompts. CARPRT introduces a training-free, class-aware reweighting scheme that adjusts the weight vector for each class based on the relevance of prompts to that class. Experiments show that CARPRT outperforms class-independent reweighting methods, highlighting the importance of modeling prompt-class dependencies.
Current prompt ensembling for VLMs uses the same weights for all classes, which is suboptimal.
CARPRT computes class-specific prompt relevance scores without additional training.
A new study enhances small language model (SLM) reasoning by grounding them in knowledge graphs via a neuro-symbolic agentic framework. Experiments on CLUTRR with Gemma 3 and Llama 3.2 show RGCN-derived hints improve performance by 1.5-2x, but reveal extraction bottlenecks and sequential deductive fragility.
Small language models (SLMs) gain reasoning boosts from knowledge graph grounding, offering a cheaper, greener alternative to LLMs.
Neuro-symbolic framework uses extract_facts and get_hint tools, leveraging RGCN for expert reasoning.
This paper addresses the 'behavioral inertia' problem in tool-augmented LLM agents when expanding their toolset. By injecting counterfactual anchor contexts at critical decision points, the proposed ToolAnchor framework breaks this inertia, recovering failed trajectories. It uses teacher models to hypothesize counterfactuals, verifies them via student rollouts, and internalizes successful interventions through post-training. Evaluated on GAIA, BrowseComp, and VDR-Bench, it shows competitive performance, bridging static post-training and dynamic adaptation.
Identifies behavioral inertia as the key obstacle in toolset expansion for LLM agents.
Proposes injecting counterfactual anchor contexts to break inertia and recover failed trajectories.
Researchers propose a novel method using Large Language Models to build Bayesian Belief Networks, employing a panel of AI agents to estimate probabilities based on personas and context, and applying a trimmed-mean rule to reduce noise. A six-step framework is illustrated on customer intention to consult a doctor in an alternative healthcare system, revealing that subjective norms have a much stronger effect than self-efficacy, and the most effective strategy is to improve both confidence and community norms simultaneously.
New method uses LLMs and a panel of AI agents to estimate probabilities, with a trimmed-mean rule to reduce noise.
A six-step BBN framework is developed for decision-making under uncertainty.
A new approach called LLM-T1D combines reinforcement learning with large language models to create an interpretable insulin pump controller for Type 1 Diabetes, achieving 73.5% Time in Range while providing clear explanations.
Combines RL with LLMs for transparent decision-making
Inspired by human communication of spatial information, language-guided geo-localization has gained traction but relies on static one-shot retrieval, failing to handle ambiguity. This paper proposes a paradigm shift to reasoning retrieval with Dialogue Place Recognition (DlgPR), which casts localization as an interactive dialogue-driven process. The authors introduce DlgQuest-Cities, the first large-scale dialogue-based benchmark for place recognition, and a unified framework with a cross-modal retriever and intelligent questioner DQ-pilot. DQ-pilot is trained via curriculum learning: supervised fine-tuning on DQ-cities-20k then reinforcement refinement on DQ-cities-10k using GRPO. Two metrics guide learning: Discriminative Difficulty Index (DDI) and Positional Retrieval Gain (PRG). Experiments show significant improvements over baselines.
Proposes Dialogue Place Recognition (DlgPR), transforming localization into an interactive dialogue reasoning process.
Introduces DlgQuest-Cities, the first large-scale dialogue-based benchmark for place recognition.
arXiv:2607.14095v1 Announce Type: new
Abstract: Retrieval Augmented Generation (RAG) has proven to be a widely successful process at improving the quality of outputs from a Large Language Model (LLM) for wider context. However, RAG systems typically retrieve context from flat document stores, which struggles when queries require hierarchical or relational reasoning across structured knowledge. I present HG-RAG (Hierarchy-Guided RAG), a framework that performs graph-traversal over a hierarchical knowledge graph to deliver structured context to a language model. My retrieval pipeline resolves a named entity anchor from the query, then expands context upward through parent nodes, laterally through relational neighbors, and downward through child nodes when needed. I evaluate HG-RAG against a dense retrieval baseline across three world scales (18-800 nodes) with four query types: local fact, hierarchical, neighborhood, and multi-hop. Results show HG-RAG consistently outperforms the flat baseline on hierarchical, relational, and multi-hop reasoning tasks, while reducing hallucination and maintaining locality coherence.
HG-RAG performs graph traversal over hierarchical knowledge graphs for retrieval.
Evaluated on three scales (18–800 nodes) and four query types.
Alphabet shares fell 4% on Thursday following a report that the company has delayed its flagship AI model, Gemini 3.5 Pro. The model's coding capabilities fell short of internal expectations, while competitors like OpenAI and Meta have released more advanced coding models.
Alphabet shares fell 4% due to delay of Gemini 3.5 Pro AI model.
Coding capabilities of the model missed internal targets; rivals have launched superior coding models.
Vision-language models (VLMs) fail to infer shared visual concepts from sets of example images. The new Visual Concept Inference from Sets (VICIS) benchmark evaluates this capability. The authors propose a training framework and architecture that learns to extract concept-specific embeddings from image sets, improving generative accuracy and generalization to unseen concepts and modalities.
VLMs cannot infer concepts from purely visual context sets, defaulting to biased generations.
VICIS task tests the ability to apply context-defined concepts to new query images.
Puter compiled Firefox to WebAssembly, enabling a full browser to run inside another browser. The project used an estimated $25,000 in Claude Opus and Fable tokens, leverages the Wisp protocol for proxying, supports end-to-end encryption, and is open source.
Puter successfully compiled Firefox's Gecko engine to WebAssembly, allowing a browser within a browser.
The project cost approximately $25,000 in AI compute resources, using a Claude Max subscription.
Artificial Analysis released AA-Briefcase agentic knowledge work benchmark results; Kimi K3 scores 1547 Elo, ranking first, surpassing GPT-5.6 Sol's 1495. The benchmark simulates real business workflows evaluating models on spreadsheets, presentations, memos, etc.
Kimi K3 ranks first on AA-Briefcase benchmark with Elo 1547.
GPT-5.6 Sol scores 1495, ranking third behind Claude Fable 5.
OpenAI's GPT-Red uses human-AI collaboration for red teaming, a novel approach to model safety, but enterprises must still ensure alignment with their workflows.
Moonshot AI released Kimi K3, a 2.8 trillion parameter model claiming to be the first 'open 3T-class model'. It outperforms many models on benchmarks but comes with higher pricing. The author tests it with a pelican-on-bicycle SVG prompt, revealing reasoning costs, hidden system prompts, and vision capabilities, while reflecting on the limitations of this informal benchmark.
Kimi K3 has 2.8 trillion parameters, is Moonshot AI's most capable model, and open weights promised by July 27, 2026.
Pricing at $3/M input and $15/M output tokens makes it the most expensive Chinese AI lab model to date.
This article describes an autonomous AI music video generation system that compares Claude Fable 5 and GPT-5.6 Sol under budgets of $25 and $100. The system lets models autonomously research, generate clips, edit, and assemble a complete video. Results show all runs produced valid videos, though quality was average, with issues in consistency and tempo matching. Claude Fable 5 was more expensive but faster, while GPT-5.6 Sol showed more creativity in editing.
System autonomously generates music videos from AI models with fixed budgets of $25 and $100.
All four runs produced complete videos, but quality remains improvable.
xAI's Grok 4.3 is now generally available on Amazon Bedrock, offering configurable reasoning effort, strong tool use, instruction following, and a 1 million token context window for agentic and enterprise workloads. This post covers its features, access methods, and how to use key capabilities such as chat, reasoning, tool calling, structured output, image input, and multi-turn conversations.
Grok 4.3 is available on Amazon Bedrock via the Mantle inference engine with OpenAI-compatible APIs.
Supports configurable reasoning effort (none, low, medium, high) to balance depth and latency.