Anthropic releases Claude Sonnet 5 and Claude Science; US lifts export restrictions on Anthropic models; Google debuts video/image generation models and Mac agent assistant; AI chip and platform startups raise significant funding; Nvidia expands data center and robotics initiatives.
On June 30, Anthropic released Claude Sonnet 5, a more affordable model optimized for running AI agents with enhanced safety features.It targets enterprise automation markets with lower pricing than the flagship Opus.
Google opened Gemini Omni Flash via its API, enabling conversational video editing.The accompanying Nano Banana 2 Lite generates 1K images in ~4 seconds at half the cost of its predecessor, with both models working together for image-to-video workflows.
MiniMax published the M3 series on Hugging Face, including a base model and MXFP8 quantized version.Using a mixture-of-experts architecture, it supports image-text-to-text tasks with built-in agent and code capabilities, quickly surpassing 570K downloads.
Om AI's VLX is designed for real-time processing of video and audio on phones and IoT devices with millisecond latency, aiming to bring AI to physical-world applications.
Anthropic launched Claude Science, a research-focused workstation integrating NVIDIA's BioNeMo Agent Toolkit for life sciences.It also initiated an AI drug discovery program, competing with Google and Nvidia in the healthcare AI space.
Google brought its 24/7 agentic assistant to macOS, adding real-time tracking and cross-app operation via natural language.This extends Google's agent ecosystem to Apple's desktop platform.
Startup Acti released an AI keyboard for iOS and Android that embeds agents into the input interface.Users can create custom shortcuts using natural language without coding, potentially changing how users interact with AI.
OpenClaw released mobile versions, bringing its open-source agent programming framework to smartphones.Users can run and interact with agent programs directly on mobile devices, lowering the barrier to agent technology.
AI chip startup Etched achieved a $5 billion valuation and announced $1 billion in contracts for its AI inference systems.The company focuses on developing Nvidia alternatives, meeting strong demand for differentiated AI hardware.
AI infrastructure firm SiliconFlow submitted its IPO prospectus, offering inference-as-a-service.Seen as a 'token factory' for the AI inference era, its listing will allow investors to evaluate the sustainability of AI inference business models.
Venice AI raised $65 million in Series A funding, reaching a $1 billion valuation.The company offers no-tracking, no-data-collection AI services and is already profitable with over $70 million in annual recurring revenue, proving the viability of privacy-first models.
The White House and Commerce Department removed export controls on Anthropic's advanced AI models, including Fable 5 and Mythos 5.Anthropic will restore user access gradually from July, signaling a major shift in US AI regulatory policy.
Infrastructure firm Firmus announced a 360MW Nvidia AI data center in Indonesia, using Nvidia's AI computing platform.It is one of the largest AI data centers in Southeast Asia, reflecting surging demand for AI compute in the region.
Nvidia joined multiple partners to invest in domestic AI manufacturing, supply chains, and energy networks, covering chip production, system integration, and workforce training.The initiative aims to reduce overseas dependency and secure AI compute resources.
Nvidia released a new open-source robot skill library, shifting from training from scratch to continual learning.Developers can incrementally add skills like grasping and placing with standardized interfaces, accelerating the path from lab to real-world applications.