Three major model releases land in 48 hours — Sonnet 5.5, GPT-6.1 Sol, and Gemini 4 Argon — all competing on cost efficiency, with Argon's 1M output token limit as the week's standout architectural leap.
Anthropic's detailed security analysis of GLM-5.3 finds frontier cyber capabilities with near-zero effective safeguards, triggering a rare 'AI security week' cluster across Anthropic, NVIDIA, and OpenAI.
OpenAI doubles the price of ChatGPT Pro while halving its limits, signalling the end of subsidised compute — the same week GPT-3's original Davinci and Babbage models are officially retired.