A Chinese lab cut the price of a frontier-class coding model to a fraction of Opus, Nvidia tried to own every layer from the laptop to the data center, and one developer ran the new Gemma 4 on a decade-old Xeon. The cost of running intelligence got attacked from both ends on the same morning — and the question underneath all of it is who gets to set that cost.

• MiniMax M3 claims parity with Opus 4.7 at roughly twelve cents per million input tokens versus five dollars — but the weights are promised in about ten days, so "open-weights" is still a countdown.
• Nvidia's DGX Station puts a GB300 chip and up to 748GB of memory on a desktop, enough to run a one-trillion-parameter model locally; the RTX Spark chip pushes the same idea into laptops, while the Vera CPUs — with Anthropic, OpenAI, and SpaceX as early customers — signal a move off x86.
• A 10-year-old Xeon is all you need: cafkafk runs a 26B mixture-of-experts model at reading speed on a 2016 CPU with no GPU, arguing mainstream tools hide the performance levers.
• Cosmos 3 is Nvidia's open physical-AI world model, backed by a Cosmos Coalition with Runway as a founding member.
• Cadence and Nvidia claim a "Level 5" autonomous chip-verification agent that turns months into a day — a large autonomy claim in a domain where mistakes ship in silicon.
• Anthropic will let the EU's ENISA join Project Glasswing for access to a model called Mythos, even as a Wirescreen analysis documents 500+ PLA attempts to procure Nvidia chips and governments from India and the UAE to France move to own their compute.