OpenAI, the Anthropic Crisis, and New AI Models

Pentagon-Focused OpenAI and Anthropic Crisis

OpenAI’s military deal with the Pentagon and the uncertainties around its limits triggered broad repercussions across the sector. In this process, Anthropic, which chose to stay independent by resisting military pressure, was classified by the US Department of Defense as a “supply chain risk.” Policy differences were also reflected in end-user metrics; while migration from ChatGPT to Claude accelerated, the Claude app reached the top in store downloads.

Next-Generation AI Models

Throughout the week, many new models focused on hardware efficiency, reasoning, and multimodal capabilities were introduced:

• OpenAI: GPT-5.3 Instant, which offers higher accuracy in daily use, and the GPT-5.4 model, accessible via API for developers and available in “Thinking” and “Pro” versions, were released.
• Google: Gemini 3.1 Flash-Lite, the fastest and most hardware-friendly model in the Gemini 3 series, was announced. The model is said to have made a major leap in logic tests compared with previous generations.
• Microsoft: Phi-4-reasoning-vision-15B, a hardware-efficient reasoning model with 15 billion parameters that can process images and text, was introduced.
• Anthropic: A new voice mode was integrated into the Claude Code interface, which runs in a terminal environment for developers.
• Alibaba: The Qwen 3.5 Small Model series, ambitious on the open-source side and stated to outperform gpt-oss-120B, was launched (parameters ranging from 0.8B to 9B).
• xAI: Grok 4.20 Beta 2 was made available, with better instruction following and reduced hallucination capacity.

Product Updates and Developer Tools

Next-generation agents and platforms were rolled out in automation and the developer ecosystem. OpenAI released the Codex app for Windows, which hosts a local sandbox, and Microsoft began developing an internal code repository as an alternative to GitHub. In cloud infrastructure, Google Cloud introduced AI agents at “Level 4” autonomy for telecom operators’ network operations; AWS announced the Amazon Connect Health platform, which automates administrative processes in the healthcare sector. In data processing, NotebookLM launched its feature that turns uploaded documents directly into animated videos.

Investment and Revenue Scaling

The market share of developer tools is growing rapidly. The annual recurring revenue (ARR) of the AI-based coding editor Cursor doubled over the past three months and exceeded $2 billion. Similarly, OpenAI’s annual revenue run rate is also estimated to have surpassed $25 billion. On the hardware and chip infrastructure side, Nvidia made a massive total investment of $4 billion in Lumentum and Coherent to accelerate optical and photonic technologies.