Anthropic researchers sharded Gemma4-26B-A4B across fifteen $55 off-brand Android TV boxes and got inference at 0.39 tokens per second, arguing that small open-weight models are months rather than years away from weight…
Meta Superintelligence Labs releases the Muse Wonder series, Muse Wonder and Muse Wonder Full, trained at its gigawatt-scale Prometheus data center and available immediately through the Meta Model API and Muse Code, bui…
Microsoft announces Maia 300, its third-generation in-house inference accelerator, designed backwards from the power envelope: 96 accelerators in a 167-kW rack, TSMC N2P compute chiplets, eight HBM4 stacks giving 288 GB…
Japan's National Cybersecurity Strategy Headquarters declares the country to be in a state of emergency in cyber space after more than twenty major firms including Lawson and Times Car disclose attacks, with the structu…