Daily AI News

OpenAI unveils GPT‑6 Astra, its first model billed as meeting the company’s AGI criteria

OpenAI announced GPT‑6 Astra, describing it as a generational leap for tasks ranging from cybersecurity to software engineering. The model is the first OpenAI release explicitly marked as meeting its internal AGI benchmark, and it will be available via the API and Cloud platform.

Read source →

Google launches WeatherNext 3, an AI‑powered weather forecasting model

Google introduced WeatherNext 3, a deep‑learning model that predicts short‑term weather with higher granularity and feeds the forecasts into Search, Maps, and Gemini. The model leverages billions of sensor readings and satellite imagery to improve accuracy over traditional numerical methods.

Read source →

Nvidia debuts RTX Spark ‘Superchip’‑powered AI laptops and mini‑PCs

At IFA 2026, Nvidia showcased the first consumer devices built around its RTX Spark Superchip, designed to run large language models locally on laptops and mini‑PCs. The hardware promises real‑time inference without cloud latency, targeting developers and power users.

Read source →

Nvidia releases PAIR, a free tool that links idle home computers into a personal AI data center

Nvidia’s Personal AI Router (PAIR) aggregates the compute of multiple idle PCs, Macs, and laptops to run local LLM inference with frameworks like Ollama and LM Studio. The software is free, open‑source‑friendly, and requires no special hardware beyond a network connection.

Read source →

Google rolls out voice‑assistant modes for Gmail, Docs, and Keep

Google added real‑time conversational voice controls—Gmail Live, Docs Live, and Keep Live—allowing users to dictate, edit, and organize content across its productivity suite. The feature leverages Gemini’s multimodal capabilities for natural, context‑aware interaction.

Read source →

Anthropic open‑sources Claude Commerce Agents, an Apache‑2.0 blueprint for shopping assistants

Anthropic released a full code‑first scaffold for building shopping and merchant agents, covering the agent loop, tool layer, approval gate, and evaluation suite. The repository, anthropics/commerce‑agents, is licensed under Apache 2.0 and ready for developers to customize across retail, travel, telecom, and entertainment.

Read source →

Perplexity open‑sources Lily, a Rust + Metal inference engine for Qwen 3.6‑35B‑A3B on Apple Silicon

Perplexity released Lily, a Rust‑based inference engine with custom Metal kernels that accelerates Qwen 3.6‑35B‑A3B on M5 Max chips. Benchmarks show 1.23× faster prefill and 1.35× faster decode compared with the MLX‑LM baseline.

Read source →

Nvidia to acquire Hugging Face for $12.93 billion

Nvidia announced a definitive agreement to purchase Hugging Face, the leading open‑source model hub, for $12.93 billion. The deal combines Nvidia’s hardware dominance with Hugging Face’s repository of models, datasets, and community tools.

Read source →

Anthropic signs $35 billion cloud‑compute deal with Lambda to scale Claude

Anthropic secured a $35 billion partnership with Lambda, an Nvidia‑backed cloud provider, to expand the compute capacity powering Claude models. The agreement includes dedicated GPU clusters and co‑development of next‑generation inference optimizations.

Read source →

Google blog details transfer‑learning methods for genomic prediction in underrepresented populations

Google researchers published a new transfer‑learning pipeline that improves polygenic risk scoring for under‑represented ancestry groups by leveraging multi‑ethnic reference panels and fine‑tuned LLM embeddings. The approach narrows prediction gaps that have hindered precision medicine.

Read source →