NVIDIA AI released Nemotron 3.5 Lightning, a 30B mixture-of-experts model with 3 billion active parameters, alongside NeMo Switchyard, a model router designed to direct agent execution steps to the cheapest capable model (MarkTechPost). The model achieved performance comparable to larger baselines while delivering nearly 670 tokens per second, prioritizing speed over maximum intelligence density (The Decoder). In parallel, NVIDIA released LTX-2.5, an open-weights world model for video generation that runs on local NVIDIA hardware, capable of generating 6.8-second clips with native multishot support and day-one ComfyUI integration (MarkTechPost).
OpenAI introduced Premium Seats for ChatGPT Business at $125 per user per month - five times the Standard Seat price - offering higher capacity and removal of five-hour usage limits, signaling a shift away from flat-rate pricing as agentic AI increases token consumption (The Decoder). Security researchers disclosed a vulnerability in reasoning-trace APIs from OpenAI, Anthropic, and Google that permits extraction of encrypted reasoning data; scanning public sessions revealed dozens of leaked passwords and API keys (The Decoder).