Showing the Nov 15 brief. Back to today ›
2 new rumors today →OfficialDomyn trains Europe's EUROPA model at ~952B parameters on the public JUPITER supercomputer · 3 days ago
Domyn published this on Nov 12, 2026 (source: domyn.com).
Domyn, lead of the EUROPA consortium with Fraunhofer IAIS and IIS, demonstrated training of EUROPA - the European Commission-backed open frontier AI model - at approximately 952 billion parameters on JUPITER, Europe's publicly funded supercomputer in Juelich, Germany, arguing Europe can compete at the frontier using existing public infrastructure. The team rebuilt the training stack on PyTorch and TorchTitan and ran 384 NVIDIA GH200 GPUs per parallel model copy, raising training throughput 14-fold from 90 to about 1,300 tokens per second per GPU to make trillion-parameter pre-training practical on public compute.
OfficialApple launches Health AI Usage Study with OpenEvidence to study trust in health AI answers
Apple Newsroom published this on Nov 15, 2026 (source: apple.com).
Apple launched the Apple Health AI Usage Study in the United States, a virtual research study and companion app conducted in collaboration with OpenEvidence. US-based participants can ask health-related questions with answers grounded in clinical data and clinical sources; the study examines what drives user trust, response quality, comprehension and everyday health actions in order to inform the responsible design of health AI experiences.
OfficialAlibaba's Qwen team puts Qwen4-Max into production API with text, image and video input
Alibaba Cloud Model Studio published this on Nov 16, 2026 (source: help.aliyun.com).
Alibaba's Qwen team moved Qwen4-Max into its production API, following the preview opened at the Apsara conference earlier this month. The flagship multimodal model now appears in Alibaba Cloud Model Studio's model catalog under complex-reasoning-and-coding and visual-understanding categories, taking text, image and video input with a 1-million-token context window. Alibaba's announcement notes internal evaluations showing roughly 62% lower average task cost in its Work360 benchmark.