foundation-model

PHASE is a physiology-guided hierarchical foundation model for intracranial EEG that uses physiological characteristics as explicit learning targets alongside masked latent prediction, setting a new state of the art on five Omni-iEEG clinical tasks and generalizing to unseen institutions.

GroundingPI：一个 4B 参数的视觉定位基础模型，以量化坐标生成点与框，在 34 个定位基准上以 73.68% 平均成绩超越 GPT-6 Astra 等 44 个基线，并在机器人操作与自动驾驶下游任务中显著提升性能。

GroundAnything 提出一个 4B 参数的视觉定位基础模型，通过块状去噪协调并行解码与精确定位，在 30 个 grounding 基准上取得同类模型新 SOTA（72.42%），并借助自投机解码实现 4.51 倍加速。

NVIDIA Kumo Tabular is an open foundation model for tabular classification and regression that uses in-context learning to achieve top performance on benchmarks without training or tuning.

The paper introduces RouteFM, a foundation model for LLM routing that learns reusable routing capabilities via episodic pretraining across heterogeneous environments, allowing a frozen router to adapt to new domains, modalities, and candidate pools through behavioral context alone. RouteFM outperforms the strongest baseline by 2.23 quality points on MMR-Bench with only eight observations per candidate, supporting a 'pretrain once, route anywhere' paradigm.

ALICE is a foundation model for zero-shot, in-context mutual information estimation, trained only on synthetic distributions and using rectified-flow velocity fields to estimate MI without per-distribution training. It achieves competitive accuracy across biology, genetics, and neuroscience benchmarks.

EXAONE Demand 1.0 is a time series foundation model designed for demand forecasting, utilizing a demand-specific corpus and adapter to outperform existing models on 22 datasets.

MM-ABC提出一个面向通用移动操作的基础模型，通过稀疏多层VLM特征实现空间感知、以世界想象与几何意图为监督的训练分支，以及MM-APT中用掩码联合注意力协调操作与移动流。模型在5000+小时、17种本体的数据上预训练，在RoboCasa365、LIBERO等基准及真实移动操作任务上取得强性能。

Dyna Robotics addresses the reliability gap in robotics by combining a large-scale data pyramid, a specialized reward model, and iterative deployment to achieve near-perfect task performance, moving towards commercial-grade autonomy.

Chamath Palihapitiya discusses his use of customized AI agents, including a Nash equilibrium agent for high-stakes negotiations and a personal AI world model for major life decisions, built as a persistent microservices system around a foundation model.

GeoSET introduces the first generalist foundation model for SAR-to-EO image translation, pretrained on over 3 million SAR–EO pairs and adaptable to downstream datasets via LoRA (0.60% parameter updates), achieving state-of-the-art FID and DISTS across six benchmarks.

The paper proposes Wiki Foundation Model (WFM), a novel foundation model for scalable, agent-native representation and retrieval using LLM Wiki, which couples dense text with link structure, achieving strong performance on benchmarks and 10.5x training acceleration.

This tweet thread claims that Brain-Computer Interface (BCI) technology will cause a 'cambrian explosion' in Human-Computer Interaction (HCI), detailing the development of a wearable BCI and the training of a brain foundation model.

This paper introduces a generalized multimodal foundation model capable of handling arbitrary modality combinations and prediction tasks, achieving competitive performance through training on large-scale synthetic datasets with diverse causal structures.

A research collaboration introduces ARLI, a method to enable RL fine-tuning of Vision-Language-Action models with asynchronous inference using real-time chunking, improving robot behaviors by allowing the RL policy to observe more recent images.

OneBid is a unified auto-bidding foundation model that addresses challenges in diverse oCPX advertising scenarios by using Mixture-of-Experts architecture and CROP optimization, with deployment at Kuaishou showing significant improvements like +13.1% in ROAS.

Cactus Needle 3 is a small, sliceable foundation model for automation tasks that runs on-device, achieving performance comparable to larger models on function calling and structured extraction.

Meta launches Muse as a standalone AI agent app, emphasizing isolation for security, but faces scrutiny over privacy and business model implications.

The paper proposes WFM, a Wiki Foundation Model for scalable, agent-native knowledge representation and retrieval, demonstrating improved performance in complex agentic reasoning tasks with a 10.5x training acceleration.

FoundAna introduces a foundation model for graph anomaly detection that integrates GNNs and transformers to achieve generalizable, cross-graph performance, demonstrating superior results on multiple benchmark datasets.