10 articles — 🔴 1 critical , 🟡 4 important , 🟢 5 interesting
Cosmos 3 is NVIDIA's new omnimodal world model released as an arXiv preprint that, within a single mixture-of-transformers architecture, simultaneously processes and generates language, images, video, audio and action sequences. The model targets embodied and physical AI and ships with open-source code, checkpoints, datasets and benchmarks.
OpenAI announced that its frontier models and Codex are now available on Amazon's cloud platform AWS. This is an expansion of the distribution of OpenAI's models and coding tools to another major cloud. Details on the exact models, pricing and regions are not elaborated in the available summary, so we point to the source.
On 1 June 2026, Google shut down the Gemini 2.0 Flash line of models in its API. Four identifiers stopped working, and developers were directed to the newer Gemini 3.5 Flash and Gemini 3.1 Flash-Lite models as a replacement for existing integrations.
OpenAI announced the construction of a new Stargate data center in the U.S. state of Michigan. This is an expansion of the Stargate compute infrastructure program, which OpenAI frames within the context of the so-called Intelligence Age. Details on capacity and investment are not elaborated in this announcement, so we point to the official source.
AMD announced that Alibaba's open-source reinforcement-learning framework ROLL now runs natively on AMD Instinct GPUs with the ROCm software, with no code changes, custom patches or non-standard builds. The collaboration includes vLLM compatibility, fixes for Ray and support for distributed RL training of large language models.
Google revealed how it used Gemini and experimental AI models to create elements of the I/O 2026 conference, from the animated short film 'TPU Training Day' to the visual brand identity. The approach combines human art with AI tools, preserving authorial intent while speeding up prototyping.
The PyTorch Foundation published a case study in which LinkedIn re-architected its distributed linear-programming solver DuaLip on PyTorch. The new GPU-accelerated version achieves up to 75x faster iterations than the original Scala code and solves optimization problems with trillions of decision variables.
