Saturday, February 7, 2026

GPT-oss-120B / GPT-oss-20B. OpenAI releases their first open-weight LLMs since GPT-2 in 2019, featuring Apache 2.0 license and training with RL and distillation from o3.

Gemini 3 Flash. Google's new model achieves Gemini 3 Pro-class reasoning at Flash-tier latency while being 3x faster and less expensive than 2.5 Pro.

NVIDIA Cosmos Reason 2. Open reasoning VLM that enables machines to see, understand, and act in the physical world, paired with Isaac GR00T N1.6.

Qwen3-Max-Thinking. Alibaba's flagship reasoning model with adaptive tool-use that intelligently invokes retrieval and code interpreter on demand.

Monty Python Interpreter. A minimal, secure Python interpreter written in Rust specifically designed for use by AI systems. Source

Understanding Neural Networks, Visually. Interactive visualization tool for comprehending how neural networks function and process information. Source

RememOry. A system designed to help users regain access to their computers if they lose their memory. Source

Step3.5-Flash in llama.cpp. Support for Step3.5-Flash model has been successfully merged into the popular llama.cpp framework.

Why I Joined OpenAI. Brendan Gregg shares his perspective on joining OpenAI and the opportunities in AI infrastructure. Source

How to Write Quality Code with AI. Comprehensive guide on effectively leveraging AI tools for software development and maintaining code quality. Source

SkyReels V3. First open-source model supporting three video generation modes in one architecture, including multi-subject reference image-to-video.

CPU-Only AI Tools. Demonstration that modern computers can run various AI tools locally without requiring GPU acceleration.

<400ms Voice Agent on GTX 1650. Community member builds ultra-low latency voice agent with hierarchical RAG running on budget 4GB VRAM GPU.

Quad 3090 "Poor Man's RTX 6000". Community build showcasing an all air-cooled system with four RTX 3090 cards as a cost-effective alternative to enterprise hardware.