LLM
- DeepSeek announces and releases (open source) DeepSeek-v3.2 and DeepSeek-v3.2-speciale, reasoning models built for agents (tech report).
AI Agents
- Google Workspace Studio enables simple creation of AI automations.
Data
- Progress on neural approaches to tabular data: Accurate predictions on small data with a tabular foundation model.
Image Synthesis
- ByteDance Seedream v4.5. Image synthesis and editing.
Video
- Odyssey announces Odyssey-2, a real-time video model that one can guide on the fly (albeit lower resolution than premier video models).
- Runway ML announces Gen 4.5 video model.
- Kling:
- Kling O1, omni model (examples, more examples).
- Kling 2.6, video generation with native audio.
- PixVerse v5.5 released; video+audio.
- Kinetix Kamo-1 focuses on controllability.
3D
- Hunyuan 3D Studio claims art-grade 3D generative model, Hunyuan 3D-PolyGen 1.5.
Robots
- Interesting to look back on the state of humanoid robots in 2015, just 10 years ago.
- UK-based Humanoid shows progress on their Alpha biped.
Cars
- Wayve touts GAIA-3 model for autonomous driving.
Science