Readable Rewrite — Improved hardware detection module for canirun-ai
Julien Chaumond (Hugging Face) has completed a rewrite of the hardware detection module for canirun-ai, focusing on readability and JSDoc documentation.
Sources x.com
Project Tapestry: Yann LeCun and AI Alliance's Plan to "Save" the AI Industry
Yann LeCun and the AI Alliance have announced Project Tapestry, a decentralized and open-source AI ecosystem aimed at breaking the monopoly of big tech corporations.
Sources x.com
DeepSeek V4 Pro Slashes Prices Permanently by 75% — Breaking the Token Monopoly
Chinese startup DeepSeek has announced a permanent 75% price cut for its flagship V4 Pro model, directly challenging major Silicon Valley labs with its cost-optimized architecture.
Sources venturebeat.com
AutoTTS: Automating Inference Strategies, Cutting LLM Token Costs by 69.5%
The new AutoTTS framework enables large language models to automatically search for optimal inference strategies, cutting token consumption by up to 69.5% while enhancing problem-solving performance.
Sources venturebeat.com
Aleph Prover formalizes the disproof of Paul Erdős's planar unit problem in Lean 4
Aleph Prover has announced the formalization of OpenAI's disproof of Paul Erdős's 'planar unit' problem. The entire source code has been open-sourced for community verification.
Sources x.com
llama.cpp b9235: Accelerating Inference with Speculative N-gram Tuning
The llama.cpp b9235 release introduces Speculative N-gram Tuning, significantly optimizing decode speeds when running large models like Qwen3.6 27B.
Sources x.com
llama.cpp Supports Multi-Token Prediction for Qwen3.6: A Quantum Leap in Performance
A new milestone for local AI as llama.cpp officially supports Multi-Token Prediction (MTP) for the Qwen3.6 series, dramatically boosting processing speeds on consumer hardware.
Sources x.com
physics-intern: Framework Helps Gemini 3.1 Pro 'Outperform' GPT 5.5 Pro in Science
A new tool called physics-intern significantly boosts the performance of large language models like Gemini 3.1 Pro in solving physics and scientific problems, thanks to a specialized subagent mechanism.
Sources x.com
Clawpatch: AI Agent Specialized in Code Review and Automated Bug Fixing
More than just a coding assistant, Clawpatch operates as a true AI Engineer, automatically scanning projects by feature and applying patches validated through existing test suites.
Sources github.com
Jim Fan open-sources CaP-X: A 'vibe agents' ecosystem for robotics
Jim Fan (NVIDIA) has announced the open-sourcing of CaP-X, an agent framework that enables robots to self-synthesize skills and interact with the physical world through perception and actuation APIs.
Sources x.com
Optimizing the SAM Model on Apple Silicon: 1.25x Increase in Inference Speed
A developer has successfully ported the SAM 2.1 (Segment Anything Model) to Apple's MLX framework, boosting performance by 25% on Mac computers.
Sources x.com
Optimized 9B Model for Tool Calling Released on Hermes Agent
A new 9-billion-parameter language model specifically trained for tool calling and agentic programming workflows has officially launched, running efficiently on consumer-grade hardware.
Sources x.com
Hugging Face CEO: Top AI Companies Are Shifting to Training Their Own Models
Observing Cursor's aggressive push to train its own models, Hugging Face's CEO notes that serious AI companies will soon abandon API reliance to return to open source.
Sources x.com
Hugging Face Partners with Dell to Boost On-Premise AI, Relieving GPU Shortage
Hugging Face's CEO believes that open-source AI running on local/on-premise infrastructure will be the solution to GPU shortages and expensive API costs.
Sources x.com
Hugging Face Launches Ettin Reranker — A Suite of 6 Fast and Accurate Ranking Models
Hugging Face has announced Ettin Reranker, a family of 6 CrossEncoder models based on ModernBERT that improves search (RAG) accuracy while running up to 8 times faster.
Sources huggingface.co
US Rumored to Tighten AI Rules: Will New Models Face a 90-Day Vetting Period?
Abacus.AI CEO Bindu Reddy warns of a new White House executive order that could force frontier AI models to undergo a 90-day government review before launch.
Sources x.com
Open-Source AI Is Accelerating in the Race to Innovate Inference Efficiency
While tech giants pour billions of dollars into massive GPU infrastructure, the open-source AI ecosystem is forced to innovate to optimize inference capabilities and achieve astonishing efficiency.
Sources x.com
Anthropic transfers Petri tool to Meridian Labs for independent development
Anthropic has decided to hand over Petri, an open-source alignment tool, to Meridian Labs, alongside a major update that enhances AI testing capabilities.
Sources x.com
Comparing the Strength of Open-Source AI Ecosystems Among Tech Superpowers
A report analyzing the differences and development trends of open-source AI ecosystems in the US, China, Europe, and other Asian nations.
Sources x.com
GoodAIList.com — A Tool for Tracking 14,000 Open-Source AI Repositories
Chip Huyen introduces GoodAIList.com, a website that automatically tracks and updates the developments of over 14,000 open-source AI repositories based on community contributions.
Sources x.com
Introducing TokenSpeed: An Open-Source LLM Inference Engine with TensorRT-Level Performance
TokenSpeed is a new LLM inference engine that matches TensorRT-LLM in performance while remaining as easy to use as vLLM, released under the MIT license.
Sources x.com
China builds practical and low-cost open-source AI models
Open-source AI models from China are becoming increasingly practical, efficiently handling 50% of daily tasks at operating costs 30 times cheaper than major competitors.
Sources x.com
TokenSpeed — New Open-Source Inference Engine Officially Launches Preview
Backed by Together AI, TokenSpeed is an MIT-licensed inference engine that promises to significantly accelerate processing for large language models.
Sources x.com
DeepSeek-V4-Pro — 1.6-Trillion Parameter Open-Source Model Officially Available on NVIDIA NIM
NVIDIA has brought DeepSeek-V4-Pro to its cloud platform, allowing users to experience the world's largest open-source model running on Blackwell chips for free.
Sources x.com
Meta launches Llama 3.1 405B — the world's most powerful open-source AI model
Mark Zuckerberg officially announced Llama 3.1 405B, the first open-source model to reach 'frontier' class, competing directly with OpenAI's GPT-4o.
Sources x.com
Hugging Face showcases open-source robotics project at AI Week LeRobot
Hugging Face is participating in the AI Week LeRobot event, marking its strong presence in the open-source robotics sector. The event features experts such as Steven Palma Morera.
Sources x.com
The Open-Source Era: When Robots Begin to 'Think'
From ROS to Hugging Face's LeRobot and Nvidia's GR00T, open-source software is lowering the barriers to entry in robotics, empowering robots with the ability to reason and make decisions.
Sources spectrum.ieee.org
GitHub honors 10 outstanding open-source projects for game developers 🎮
GitHub has announced a list of 10 open-source projects that help optimize game production workflows, ranging from graphics and audio processing to dialogue systems.
Sources github.blog
Hugging Face Champions Open Source for AI Security
Hugging Face highlights the role of transparency and open source in the future of AI security, enabling the community to detect and patch vulnerabilities faster.
Sources huggingface.co
Top Free Open-Source BI Tools for 2026
Analytics Insight compiles a list of powerful open-source Business Intelligence (BI) platforms, helping businesses optimize data analytics costs in 2026.
Sources analyticsinsight.net
NVIDIA Cosmos 3 Tops Artificial Analysis Leaderboards for Open Weights Models
NVIDIA's Cosmos 3 has secured the #1 spot in both Text-to-Image and Image-to-Video categories on the Artificial Analysis leaderboards. It is a family of omnimodal world models for Physical AI, unifying language, image, video, audio, and action.
Sources x.com
Nvidia's Nemotron 3 Ultra leads US open models
Nvidia's new Nemotron 3 Ultra has achieved the highest performance among open-source AI models from the US, according to Artificial Analysis.
Sources the-decoder.com
Hugging Face CEO Urges Public Sharing of Agent Traces for Open Source AI
Hugging Face CEO Clement Delangue is advocating for the public sharing of coding and agent traces to build better open-source datasets and models.
Sources x.com
Maxime Labonne's 1B Model Trends on Hugging Face for Agentic Power
Despite having only 1B parameters, Maxime Labonne's new model is trending on Hugging Face for its surprisingly high performance on agentic tasks.
Sources x.com
Claw Code — A Rust-based harness for CLI agents
An open-source Rust implementation of the `claw` tool that allows running AI agents via CLI, supporting Anthropic and OpenAI APIs instead of relying on a Claude subscription.
Sources github.com
OpenClaw — A Multi-Platform Personal AI Assistant Gateway
The open-source OpenClaw project allows users to run personal AI assistants across multiple channels such as WhatsApp, Telegram, and Discord, featuring voice control and an intuitive Canvas.
Sources github.com
Edalitics: An easy-to-use platform for data analysis and dashboard creation
Edalitics is an open-source data analytics and dashboard platform that makes it easy for anyone, including non-technical users, to explore and visualize data.
Sources github.com
GoatCounter — Open Web Analytics, Worry-Free Data
GoatCounter is a simple, privacy-focused, open-source web analytics platform that can be used as a free hosted service or self-hosted.
Sources github.com
Files.md – A private, minimalist Markdown note-taking space for you
Files.md is a minimalist, privacy-focused, and local-first note-taking app that helps you manage Markdown files efficiently.
Sources github.com
CLI-Anything: Connecting AI Agents to Any Software via Command Line
CLI-Anything is a platform that helps AI agents control legacy software through standardized command-line interfaces, turning any application into an 'agent-native' tool.
Sources github.com
ClimateExplorer: An Open-Source Platform for Exploring Climate Change Data
ClimateExplorer provides an intuitive web interface that helps users track local and global climate change indicators from reliable data sources.
Sources github.com
Llama.cpp Launches Official Homepage, Optimizing Local AI Experience
Llama.cpp has officially launched the llama.app website alongside a cross-platform installer that runs via a single command, making local AI more accessible than ever.
Sources x.com
Lotto Labs launches 8B model optimized for consumer hardware (GPU-poor)
Lotto Labs has introduced an 8-billion parameter (8B) AI model trained on a massive number of tokens. This model is specifically designed to run ultra-fast on modest hardware, bringing hope to the "GPU-poor" community.
Sources x.com
Release of stable-worldmodel — an open-source platform for JEPA and World Model research
After a year of development, stable-worldmodel has been officially launched. It is an open-source, scalable platform designed to accelerate AI research in JEPA and World Models.
Sources x.com
Repo2RLEnv: Turning Any Repository Into an AI Coding Agent Training Environment
Repo2RLEnv is an open-source tool that allows converting any repository into a runnable and verifiable coding environment. Based on real-world PRs and commits, this tool provides powerful support for evaluating and training AI models (Reinforcement Learning) in the software engineering domain.
Sources x.com
AI: EvolutionaryScale Launches World Model for Protein Biology: ESMFold2
EvolutionaryScale announces the first world model for protein biology, providing a powerful tool for scientific prediction, design, and discovery.
Sources x.com
AI: EvolutionaryScale Open-Sources Entire Suite of Protein Design Models and Protocols
EvolutionaryScale has open-sourced its entire suite of ESM models and binder design protocols to support the protein biology research community.
Sources x.com
EvolutionaryScale Launches ESMC: An AI "Map" Trained on 2.8 Billion Protein Sequences 🧬
EvolutionaryScale has announced ESMC, an open-source protein language model trained on 2.8 billion protein sequences from across Earth's entire ecosystem.
Sources x.com