Digest · past 24h

1. Thinking Machines Releases Open Multimodal Model Inkling with 1 Trillion Parameters
   Thinking Machines has released Inkling, an open multimodal LLM with ~1T total parameters and 41B active parameters, supporting text, image, and audio inputs with a 1M context window. The model is available on Hugging Face in BF16 and NVFP4 variants, with day-0 support in transformers, SGLang, vLLM, and llama.cpp.
2. AMD Unveils Helios with World's First 2nm GPU MI455X: OpenAI, Meta, and Microsoft Among Customers
   AMD has launched the Helios rack-scale system featuring the Instinct MI455X accelerator, claimed to be the first GPU with 2nm chiplets. Customers include OpenAI, Meta, Microsoft, and Oracle, with shipments expected from Q3 2026. The system uses an open architecture with UALink and Ultra Ethernet, aiming to challenge Nvidia's dominance.
3. Google DeepMind introduces Gemini Omni: multimodal model for video creation and editing
   Google DeepMind has unveiled Gemini Omni, a multimodal AI model capable of generating and editing videos from any combination of images, audio, video, and text. The first model in the family, Gemini Omni Flash, is rolling out to Google AI Plus, Pro, and Ultra subscribers via the Gemini app and Google Flow, and will be available to YouTube Shorts users at no cost.
4. Mistral AI unveils Mistral 3 model family, including Mistral Large 3
   Mistral AI announces the Mistral 3 family, featuring three dense models (14B, 8B, 3B) and the flagship Mistral Large 3, a sparse MoE model with 41B active and 675B total parameters. All models are released under Apache 2.0. Mistral Large 3 achieves frontier performance, image understanding, and multilingual capabilities, ranking #2 among OSS non-reasoning models on LMArena.