AI News Of The Week (3rd July, 2026)
Anthropic Reopens Fable 5 and Mythos 5 After Washington Lifts Export Curbs
On June 30, Anthropic said the U.S. Commerce Department lifted export controls on Fable 5 and Mythos 5 less than three weeks after forcing the company to shut them off for all users. Fable 5 returned globally on July 1, Mythos 5 access widened again for approved U.S. organizations, and Anthropic said it would work with government and Glasswing partners on common standards for rating and fixing jailbreaks.
Claude Sonnet 5 Arrives and Shrinks the Gap to Opus-Class Agents
Also on June 30, Anthropic introduced Claude Sonnet 5 as its most agentic Sonnet model yet, saying it comes close to Opus 4.8 on reasoning, tool use, coding and knowledge work while staying far cheaper. It became the default model for Free and Pro plans immediately, launched across Claude Code and the Claude Platform, and gave Anthropic a much stronger mid-tier option for large-scale agent workflows.
Claude Science Turns Anthropic's Research Push Into a Real Workbench
On June 30, Anthropic also launched Claude Science, a beta app for macOS and Linux that packages scientific tools, computing access and auditable outputs into one research workspace. The company paired the launch with up to 50 AI-for-Science projects, offering as much as $30,000 in credits per project plus extra compute support, signaling a more serious move into biology and academic research workflows.
Google Pushes Gemini Spark Onto the Mac and Deeper Into Everyday Apps
On June 30, Google expanded Gemini Spark with a macOS beta, new app integrations, support for custom MCP connections and real-time topic tracking. Spark can now work across desktop files, Google Workspace and connected services like Canva, Dropbox, OpenTable and Zillow Rentals, pushing Google's agent from a chat surface toward a system that can watch, organize and act across the tools people already use.
Google Starts Shipping Gemini Omni Flash to Developers
Also on June 30, Google began rolling Gemini Omni Flash out through the Gemini API and Google AI Studio, turning its multimodal video generation and editing model into a developer product. The release supports conversational editing across text, image and video inputs, and it shows Google moving quickly to commercialize the model layer behind the agentic and creative demos it showed earlier at I/O.
Microsoft Launches Frontier Company to Help Enterprises Choose Their AI Stack
On July 2, Microsoft said it was creating Microsoft Frontier Company, a new business backed by $2.5 billion to help customers pick, adapt and deploy AI systems that actually fit their data and goals. The move matters because it openly reflects a new enterprise reality: large companies no longer want to bet everything on one model vendor, and Microsoft now wants to profit from that multi-model shift.
China's GLM-5.2 Shows the Frontier Race Is Getting Cheaper
On July 2, Reuters reported that Beijing-based Z.ai had started winning attention far beyond China with GLM-5.2, a low-cost model that analysts said sits just below Anthropic's Opus 4.8 and alongside leading GPT systems on several coding-heavy measures. Its rise matters not only because of performance, but because it suggests open and lower-cost challengers are starting to pressure the economics of closed U.S. frontier models.
