Models

Researchers train Iris search agents using web hyperlink structures: Researchers have introduced Iris-mini and Iris-pro, two search agents trained at the 35B-A3B and 397B-A17B scales. The models were trained using a data pipeline that reverse-constructs multi-hop question chains from the hyperlink structure of a web corpus.

AI Models • arXiv

Saudi firm HUMAIN builds Arabic AI model on MiniMax open stack: The PIF-backed company HUMAIN has launched an Arabic frontier AI model built on MiniMax's open Mixture-of-Experts stack. This development highlights how sovereign AI projects are increasingly utilizing Chinese open-weight models as foundational bases for national initiatives.

AI Models • Pandaily

Fields Medalist helps build hybrid mobile-cloud AI system for ARC-AGI: A new hybrid AI system combining a local 4B Qwen model on a mobile device with a cloud-based GLM model has been developed to tackle the ARC-AGI benchmark. The project involves collaboration with a Fields Medalist.

AI Models • 量子位

AI agents

US retail investors use AI agents to automate stock trading

OpenAI launches GPT-6 Astra for top-tier ChatGPT subscribers: OpenAI has launched its GPT-6 Astra model for Pro, Enterprise, and Business Premium subscribers, with Plus users expected to gain access soon. Message allowances for Astra are roughly half of those for GPT-5.6 Sol, ranging from an estimated 5 to 45 messages per five hours for Plus users.

AI Models • The Decoder

Alibaba releases Qwen3.8-Flash-Next weights for agentic coding: Alibaba has released the model weights for Qwen3.8-Flash-Next as a preview of its upcoming Qwen4 architecture. The release allows developers to experiment with and evaluate the model, specifically for agentic coding tasks on NVIDIA GB300 NVL72 systems.

AI Models • NVIDIA Generative AI

NVIDIA Jetson

Nvidia details how to deploy large reasoning models on Jetson edge devices

AI Agents

Developers use llms.txt files to optimize websites for AI agents

Meta offers 95% Muse Spark discount in exchange for user prompt data: Meta is offering an average 95% discount on its new Muse Spark model, which is designed for operating coding and other agents. To receive the discount, users must agree to share their prompts and model outputs to help Meta develop future AI models.

AI Models • TechCrunch AI

Researchers propose decision framework for proactive AI service agents: A new survey paper defines proactive service agents that infer service opportunities from incomplete environmental and user signals rather than waiting for explicit instructions. The researchers formulate this problem as a partially observable sequential decision process constrained by authorization and risk.

AI agents • arXiv

Hugging Face • Nvidia

Nvidia agrees to acquire Hugging Face for $12.93 billion: Nvidia has agreed to acquire Hugging Face, an online hosting platform for open-source AI models, datasets, and tools, for $12.93 billion. Founded in 2016, Hugging Face provides a collaborative space for AI developers to share projects and data.