ABOUT THIS FEED
AIModels.fyi is a curated resource for tracking the release of new AI models, tools, and datasets. Its Substack RSS feed provides digest-style newsletters that summarize the latest in machine learning research and open-source projects. The platform focuses on making it easier for practitioners and enthusiasts to discover and follow emerging models across natural language processing, computer vision, generative AI, and reinforcement learning. Each post aggregates relevant updates, links, and context, saving readers time compared to browsing multiple sources. The writing is concise yet informative, appealing to developers, researchers, and students who want a quick overview of cutting-edge developments. With a few posts per week, the feed is highly practical for staying on top of the rapidly expanding AI ecosystem without information overload.
- Xiaomi just open-sourced a 1T-parameter model and almost nobody noticed
MiMo-V2.5-Pro matches frontier coders on benchmarks, ships under MIT, and burns 40-60% fewer tokens per agent run - but it's 1.02 trillion parameters of MoE and you can't run it on your gaming rig.
- The prompt isn't hiding inside the image
CLIP Interrogator is one of the most misunderstood tools in the Stable Diffusion ecosystem. It solves a real problem, which is why it won't go away.
- Google’s Best Open Model Yet Has a Memory Problem
Gemma 4 31B's 256K context window is real. So is the VRAM bill that comes with it.
- Meta traded its biggest community asset for a commerce engine
Muse Spark is competitive. The open-source bet that got Meta here is over.
- Netflix's VOID shows video editing has finally learned the laws of physics
By treating object removal as a causal simulation rather than a pixel-patching job, VOID eliminates "ghost" physics from edited scenes
- Why wait until the end to realize your model’s code won’t actually run?
Think Anywhere in Code Generation
- Why pay for proprietary search APIs when you can synthesize research agents offline?
OpenResearcher: A Fully Open Pipeline for Long-Horizon Deep Research Trajectory Synthesis
- Can you *really* train AI to "get" videos just by showing it a million of them?
A Very Big Video Reasoning Suite
- Can AI *really* research like us? This new framework puts it to the test.
DeepResearchEval: An Automated Framework for Deep Research Task Construction and Agentic Evaluation
- Can an AI *finally* react like a real person during a video call?
Avatar Forcing: Real-Time Interactive Head Avatar Generation for Natural Conversation