A new open-source CLI tool called 'whichllm' instantly scans your hardware and recommends the best local AI models you can actually run, ranked by real benchmark data from top leaderboards with a single command.
Unsloth Studio launches in beta, enabling users to train over 500 open-source AI models locally on Windows, Linux, and macOS up to 2x faster while using 70% less VRAM, with features including reinforcement learning, multi-GPU support, and API inference compatibility with tools like Claude Code.
A new 1B parameter AI model called HRM-Text is making foundation model pretraining accessible to nearly anyone, requiring up to 900x less data and 600x less compute than traditional methods, bringing full pretraining costs down to roughly $1000 using a hierarchical recurrent architecture with strong benchmark performance across GSM8k, MATH, …
NVlabs launches SANA, a groundbreaking open-source AI framework capable of generating stunning 4K images and video up to 100 times faster than competitors like Flux-12B, with multiple model variants, flexible deployment options including low-VRAM laptop support, and a fully open-sourced codebase under the Apache 2.0 license.
OpenAI quietly releases two powerful open-weight AI models, GPT-OSS-120B and GPT-OSS-20B, just days before GPT-5 launches, offering state-of-the-art reasoning capabilities under a permissive Apache 2.0 license — with the smaller model running on a standard laptop with just 16GB of RAM.
A massively popular 47.6K-star GitHub repository called Made With ML is teaching developers the complete machine learning lifecycle—from training and tuning to production deployment—using MLOps best practices, MLflow experiment tracking, Ray-powered cloud scaling, and fully automated CI/CD pipelines via GitHub Actions.
Datadog launches Toto 2.0, a groundbreaking family of open-weights time series forecasting models scaling up to 2.5 billion parameters, achieving state-of-the-art results across all major benchmarks while being up to 7 times more parameter-efficient than its predecessor.
A new open-source AI voice translator called PuriPuly is revolutionizing real-time two-way communication in VRChat, supporting slang, context memory, and VR subtitles at near-zero cost with free credits for new users.
SakanaAI releases 'sparser-faster-llms,' an open-source repository of custom CUDA kernels for H100 GPUs that use sparsity to dramatically speed up LLM inference and cut memory usage, with pretrained models from 0.5B to 2B parameters now available on Hugging Face Hub.
TabPFN launches as an open-source foundation model for tabular data, supporting classification and regression tasks with both local PyTorch and cloud-based inference, while enterprise users gain access to large-scale data support handling up to 10 million rows.