October 23, 2026
Quick hits from across the AI world in the last 24 hours.
- ⚡ Quick hits
- Industry Cohere’s Model Vault entered public beta, combining confidential computing with Azure GPUs to keep enterprise model workloads private. X / Cohere ↗
- Funding Pathos AI raised $237 million to advance AI-driven drug discovery and development. SiliconANGLE ↗
- AIGC Block developed Dessa, an AI voice model designed to converse with users in an intentionally awkward, human-like style. The Verge ↗
- Agents AWS added hard monthly spending limits to its new developer experience, automatically pausing projects when they reach their caps. X / AWS Developers ↗
- Research Google Research is deploying open foundation models and clinical partnerships to improve tuberculosis screening in underserved regions. Google Research ↗
- Research Epoch AI published four benchmark reviews, rating one as verified, three as flawed, and several others inconclusive. X / Epoch AI Research ↗
- Agents Lovart introduced Delegated Tasks, letting its Arti design agent perform recurring creative work on a user-defined schedule. X / Lovart ↗
- Agents Baseten launched an Environment Network connecting reinforcement-learning environment providers for custom agent training. X / Baseten ↗
- Open Source Tencent Hunyuan released Hy-Nomas, an open-source interactive video world model controlled by text prompts and camera trajectories. X / Tencent Hunyuan ↗
- Industry OpenAI detailed how its storage platform serves more than one billion ChatGPT users through partitioning, caching, replication, and Rust services. OpenAI ↗
- AIGC HeyGen said its Video 1 model can now generate 30-second text-to-video clips in a single take at $0.01 per second. X / HeyGen ↗
- Agents Cognition released a refreshed Devin Desktop app with a redesigned command center and free downloads for ChatGPT subscribers. X / Cognition ↗
- AIGC Krea launched an integration that lets users generate images and videos with multiple models directly inside Grok. X / Krea AI ↗
- Open Source NVIDIA released Nemotron 3.5 Bolt E4B as an open, customizable small vision-language model for local tool use and assistants. X / NVIDIA AI ↗