Created by ResponsibleAI
Real‑time AI model announcements and capability overviews from leading labs and open‑source projects
Explore the latest content tracked by AI Model Pulse
Flux 3 introduces Self-Flow, a unified multimodal architecture that jointly learns image, video, and audio generation. It produces up to 20-second clips with native synchronized sound in a single pass and extends to action prediction.
Gemini Robotics ER 2 acts as a high-level planner that reads continuous video, holds conversations, decomposes tasks, and triggers external tools...
ByteDance and MiniMax launched rival AI video models on July 31 with opposite strategies.
Thinking Machines' Inkling Small underscores the push toward smaller, competitive open-source models that cut costs without major trade-offs.
Researchers at the Chinese Academy of Sciences introduced PQFA, a hybrid quantum-classical framework that slashes augmentation parameters from 24,000...
MiniMax positions its new H3 model as a cost-effective, multimodal contender against ByteDance and Kuaishou.
Amazon is pruning its sprawling Nova lineup—winding down Premier, Omni, Canvas, and Reel—to concentrate resources on a single next-generation...
Qwen3.7 Flash delivers a 1M-token context window and multimodal support at aggressive pricing.
Black Forest Labs released FLUX 3, a single-architecture multimodal model that generates video with native audio while powering robotic action...
Moonshot's official announcement celebrates Kimi K3 as a 2.8T MoE model with 1M context, 104B active params, and 2.5x efficiency gains.
HOPPR has launched the HOPPR EF Chest CT Narrative Model, a domain-specific foundation model that ingests 3D chest CT volumes and produces narrative outputs. This specialized release directly targets radiology report generation workflows.
Microsoft's MAI-Cyber-1-Flash model, integrated with GPT-5.4, scored 95.95% on the CyberGym benchmark, marking a major advance in AI-driven...
NVIDIA Cosmos models have reached 10 million downloads on Hugging Face, establishing Cosmos 3 as the world's first fully open world foundation model for physical AI.
Elastic now integrates Jina's multimodal and multilingual embedding and reranker models, delivering frontier-grade accuracy directly in its platform with zero external dependencies.
No significant updates today.
Google's Gemma 4 family puts capable local AI within reach of everyday hardware.
Black Forest Labs' FLUX 3 uses Self-Flow to train one model jointly on images, video, and audio, creating mutual constraints that improve world...
