This was a week of major model launches paired with unsettled questions about what those models can actually do, as Anthropic and OpenAI both shipped new flagship systems and Nvidia's Hugging Face acquisition moved from reported agreement to official confirmation.
Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1, positioning the new Fable tier as cheaper and less restrictive than its predecessor while keeping Mythos as the higher-capability option; the company also published guidance on building commerce agents with Claude, signaling a push into agentic shopping and transaction workflows. OpenAI countered with Astra, a powerful new model that drew immediate scrutiny: reporting found it strikingly effective at breaking into computer systems, raising the kind of dual-use concerns that have trailed OpenAI's most capable releases before. The unease deepened with word that OpenAI's newest reasoning technique has alarmed AI safety experts, suggesting both Astra's capabilities and the training approach behind it are drawing pushback from the community OpenAI relies on for credibility.
Nvidia confirmed its $12.9 billion purchase of Hugging Face, turning last week's reported agreement into an official deal and cementing Nvidia's move from chips into the developer-tooling layer. Elsewhere in the competitive landscape, Meta released Muse Spark 1.3 to narrow the gap with OpenAI and Anthropic, Google began upgrading the mobile Assistant experience to run on Gemini, and the Pentagon confirmed it now runs its own internal versions of both ChatGPT and Grok — a sign that frontier-model deployment inside government is no longer hypothetical.
Rounding out the week: Apple's Tim Cook signaled in a parting message that the company's AI direction is now in the hands of "a product builder," Nvidia's $3.5 billion MediaTek investment revealed more of its chip strategy, Anthropic published new alignment and security research, Microsoft argued that healthcare AI outcomes hinge on data-foundation work, and the U.S. government sided with OpenAI in an ongoing dispute over training on copyrighted material.
To watch next week: whether the safety concerns around Astra's reasoning approach prompt any response from OpenAI, and how quickly Anthropic's commerce-agent push turns into real product integrations.