Live intelligence

Latest AI updates

14 updates
Products

AWS launches SageMaker JumpStart Model Leaderboard for comparing foundation models

AWS has introduced the SageMaker JumpStart Model Leaderboard, a view inside SageMaker Studio that compares open-weight foundation models on quality, speed and cost. It reports AWS-run scores on BIG-Bench Hard, GPQA Diamond and MMLU-Pro, plus measured latency, throughput and instance cost for each deployment configuration. At launch the leaderboard covers 70+ models from providers including Meta, Google, Qwen, IBM and Arcee. Users can filter, sort, and deploy a model matching the configuration they compared.

Amazon Web ServicesAmazon SageMakerfoundation models
Models

This smell startup says its AI nose is as good as a human’s

Osmo, an AI startup spun out of Google Brain, says its foundation model for smell has reached “human-level olfactory intelligence” in detecting, discriminating, and creating scents. The company has built a dataset of more than 10 million human labels and 800 million scent tokens, and says its technology is already used in fragrance design and counterfeit detection. The article also notes skepticism from smell scientists who question Osmo’s marketing claims and say human sensory panels remain central to its patents.

Osmoolfactory AIfoundation models
Research Editor selected

Mantis 2 Combines Simulated and Real Data for Universal Disease Forecasting

Mantis 2 is a foundation model for infectious disease forecasting that blends simulated outbreak data with observational training data. It can jointly forecast up to six variables sixteen weeks ahead and outperformed machine learning, statistical, and baseline models across 30 forecasting tasks. The model missed roughly 30% fewer outbreak peaks than the strongest comparison model, with anticipated peaks identified more than six weeks in advance on average. The authors publicly released Mantis 2 so agencies can forecast disease outcomes without building separate models per disease or location.

Why it matters: This is a concrete, recent research advance with a released model that directly improves practical epidemic forecasting for public health agencies.

Mantis 2infectious disease forecastingfoundation models
Research

Matched-Pair Study Finds Earth-Observation Pretraining Is Not Uniformly Better

A new arXiv study compares five matched pairs of Earth-observation and generic vision encoders across 14 frozen models and five RGB scene-classification datasets. EO pretraining significantly helped SatDINO, Scale-MAE and RemoteCLIP, but significantly hurt DINOv3-SAT and SatMAE on linear probing, with DINOv3-SAT dropping by up to 22.5 points. Two MAEs pretrained on the same fMoW-RGB imagery produced opposite outcomes, and gains or losses were largest with 1-5% labels. The authors conclude that satellite pretraining is not a reliable selection criterion and that candidate checkpoints should be validated on target data.

Earth observationfoundation modelsrepresentation learning
Research

DINOv2 Adaptation Generalizes Better Than Task-Specific Medical Segmentation Models

A new arXiv study compares adapting the DINOv2 foundation model with training U-Nets and ResU-Nets from scratch for medical image segmentation across colonoscopy, breast ultrasound and dermoscopy. The adapted foundation model generalized better on every independent external dataset, while task-specific models remained competitive in-distribution when training data were abundant and delineated boundaries more precisely. Only a small fraction of the foundation model needed adaptation, with the required capacity determined by the imaging domain rather than the tuning method; the authors also found that a widely used polyp test set duplicates images from another benchmark.

Medical image segmentationFoundation modelsDINOv2
Research

InnerLogic-Surg: An Open Surgical Foundation Model via Continual Pretraining

Researchers present InnerLogic-Surg, an open family of surgical vision encoders (21M–300M parameters) obtained by continually pretraining a transformer on millions of surgical images from nearly 3,000 hours of video spanning over fifteen procedure types. The method inherits DINOv3's pretraining stage and adds a learned feature-upsampling module for efficient high-resolution dense prediction. Across 20 surgical tasks including workflow, instrument and anatomy recognition, segmentation, detection, and depth estimation, the largest model achieves the best linear-probe performance under the same architecture. The work demonstrates that surgical continual pretraining yields transferable representations that generalize across procedures and task types, offering a scalable alternative to procedure-specific approaches.

surgical computer visionfoundation modelsmedical AI
Research Editor selected

Active In-Context Learning for Tabular Foundation Models: When, Where, How

This paper studies active in-context learning for tabular foundation models (TFMs), where labels are acquired iteratively to build the context. The authors propose and evaluate target-aware and information-gain-based acquisition rules against classical methods across twenty benchmarks. Key findings show that diversity-based acquisition works best early, target-aware nearest-neighbor selection is crucial, and the EPIC rule retains 80-87% of full-data performance with only 5-10% of the pool labeled.

Why it matters: Provides concrete guidance on label-efficient strategies for tabular foundation models, a practical concern for real-world deployment.

active learningtabular datafoundation models
Research Editor selected

Full-Stack Survey of Dexterous Robotic Hands in the Foundation-Model Era

This survey reviews dexterous robotic hand research from January 2024 to early November 2026, organizing the field into five layers: hardware, data, cross-embodiment transfer, learning, and evaluation. It highlights that human data value depends on kinematic, visual, and contact alignment, and that foundation policies still face embodiment-specific constraints. The paper identifies eleven open problems and proposes reporting standards for more measurable progress.

Why it matters: Provides a structured, current synthesis of a rapidly evolving area, offering actionable insights for researchers and practitioners.

robotic manipulationfoundation modelsembodiment
Research

TailSteer: Trajectory-Tail Steering for Drift-Resistant Multi-turn Image Editing

TailSteer is a final-stage training framework that improves multi-turn image editing robustness by concentrating preservation corrections near the trajectory tail of the denoising process. It learns from the editor's own on-policy rollouts using unchanged source content as reference, avoiding curated ground-truth pairs. Experiments on FLUX, SenseNova, and Qwen families show reduced cumulative drift over 15-20 turns while preserving base-model performance.

image editingmulti-turn robustnessfoundation models
Research

Why Tabular Foundation Models Excel on Relational Data They Have Never Seen

This paper investigates why single-table tabular foundation models (TFMs) perform well on relational databases despite no relational pre-training. The authors confirm this phenomenon empirically and provide multiple complementary explanations based on parameter-free relational encoders that map relational effects to embeddings mimicking TFM training distributions. The analysis suggests new directions for improving relational database foundation models, especially for interpretability and agentic integration.

Foundation ModelsRelational DataIn-Context Learning
Research Editor selected

scAnyOmics: Foundation-Model Distillation for Any-to-Any Completion of Multi-omics Virtual Cells

scAnyOmics is a generative framework that distills knowledge from a frozen single-cell foundation model into a multimodal student model, enabling cross-modal prediction and reconstruction of missing modalities from scRNA-seq, protein-marker measurements, and scATAC-seq. Evaluated on 690,880 cells from six studies, it outperforms baseline methods on held-out donor data. Applications include cell-type annotation, perturbation-response prediction, and biological age estimation.

Why it matters: Offers a concrete, validated method for integrating incomplete multi-omics data using foundation model distillation, with clear performance gains and practical applications.

single-cell genomicsmulti-omics integrationfoundation models
Research Editor selected

General-purpose foundation models outperform specialized chemical FMs on property prediction with scarce data

A new arXiv study compares specialized chemical foundation models (ChemBERTa, MoLFormer) with general-purpose models (GPT-6 Astra, Claude Opus 5.5) on chemical property prediction tasks. With only 60 experimental datapoints, GPT-6 Astra outperforms the chemical FMs on five regression tasks. Chemical FMs only match Astra's performance when given 10x more data, suggesting general-purpose models are better for data-scarce chemistry research.

Why it matters: This finding challenges the common assumption that domain-specific models are superior in specialized fields, offering practical guidance for researchers in chemistry and AI.

foundation modelschemistryproperty prediction
Research Editor selected

Chronos-2: A Foundation Model for Multivariate Time Series Forecasting

Chronos-2 is a new foundation model for zero-shot forecasting that models interactions among multivariate time series and heterogeneous covariates, unlike prior models that treat series independently. It achieves state-of-the-art accuracy on average, improving skill scores by 5.2 percentage points and win rates by 10 percentage points over strong baselines. The model is computationally efficient, generating forecasts for 100 series in under a second on a mid-range GPU, and matches or surpasses specialized methods in domains like infectious disease, macroeconomics, carbon flux, and runoff.

Why it matters: This introduces a significant advance in time series foundation models, enabling multivariate interaction modeling and broad cross-domain applicability with strong performance and efficiency.

Chronos-2time series forecastingfoundation models
Research Editor selected

Timestral open time series models claim top spots on forecasting benchmarks

Timestral is a family of four probabilistic time series foundation models with 5M, 28M, 123M, and 209M parameters. The design uses Decoupled Horizon Forecasting to separate prediction tokens from causally encoded context, a covariate framework for known-future covariates, and a two-stage pipeline that pretrains only on large-scale synthetic data before post-training on mixed real and synthetic data with longer context. The report says Timestral-209M and Timestral-123M rank first and second on every metric of GIFT-Eval, TIME, and fev-bench. Weights for all four models are on Hugging Face, with inference code on GitHub.

Why it matters: Open model weights plus a claimed new state of the art on major time series benchmarks give researchers and practitioners a concrete forecasting advance to evaluate.

Timestraltime series forecastingfoundation models