/
Published on 27.03.2026
TLDR: Nvidia released Nemotron 3 Super 120B-A12B, a fully open-weights language model with a hybrid mamba-2/transformer/mixture-of-experts architecture that pushes 442 output tokens per second, making it the fastest open-weights model in its class and a clear play to keep the developer ecosystem locked to Nvidia hardware.
The Batch - Nvidia Nemotron 3 Super 120B
TLDR: OpenAI and Amazon announced a stateful runtime environment for AI agents on AWS, giving Amazon exclusive third-party cloud hosting rights for OpenAI's frontier models and further unwinding the once-central Microsoft partnership.
The Batch - OpenAI Agent States on AWS
TLDR: xAI launched Grok Imagine 1.0, a text/image/video-to-video generator that topped Artificial Analysis Video Arena while costing roughly a third of Google Veo 3.1 and a seventh of Sora 2 Pro per minute of generated video.
The Batch - xAI Grok Imagine 1.0
TLDR: MIT researchers developed Recursive Language Models that process long contexts by treating input text as a persistent variable in an external Python environment, recursively breaking tasks into sub-tasks and achieving 91.3 percent on BrowseComp+ where GPT-5 alone could not answer.