Friday 6 November 2026
Google DeepMind releases Gemini Omni 1.2 Flash for faster, cheaper video with longer clips
Google DeepMind leads the day with a faster, cheaper Gemini Omni video model. NVIDIA and Tencent also shipped new models, while SemiAnalysis launched GDPx, a benchmark where no AI agent scores above 12 percent.
Google DeepMind releases Gemini Omni 1.2 Flash for faster, cheaper video with longer clips
Google DeepMind adds a fast, cost-effective video option that supports longer clips, widening access to its Gemini Omni line.
Inkling 2 Small released
Models: Inkling
NVIDIA releases Cosmos 3 open models for interactive world simulation
NVIDIA's open Cosmos 3 models target interactive world simulation, with the release shared on the Hugging Face blog.
Tencent releases Hy4, its first multimodal model with vision and computer use
Tencent's first multimodal model with vision and computer use scores 35 on the Artificial Analysis index, though Tencent admits it trails top models on hard math.
SemiAnalysis launches GDPx benchmark where no AI agent scores above 12 percent
No agent tops 12 percent on simulated knowledge work, with Claude Opus 5.5 narrowly ahead of Gemini 4 Carbon; critics question task counts and testing conditions.
Models: Gemini 4 Carbon, Claude Opus 5.5
WeChat unveils WeLM-Image, a unified image generation and editing model
WeChat's research model unifies image generation and editing for single-GPU use, with a Turbo variant making a 1024x1024 image in 4.98 seconds on one H20, per the paper.
Z.ai chief scientist outlines training pipeline for upcoming GLM-5.5 models
Unconfirmed: Z.ai chief scientist Jie Tang reportedly described how GLM-5.5 is trained, with a 32B model among planned smaller sizes.
Meta adds real-time voice conversations to its Muse AI agent
Muse users can now talk to the agent by voice with a lip-synced avatar and adjustable tone, though it sometimes pauses while looking things up.
Written by Claude (claude-opus-5-5) from the stories below.