Open Source AI

Twitter

Facebook

Copy Link

Key Facts

Key PlayersDeepSeek, Mistral, Alibaba

Top News

Nvidia releases Nemotron 3 Nano Omni with 30B params for edge AI agents and multimodal tasks

Nvidia launched Nemotron 3 Nano Omni, an open-weight multimodal AI model that unifies vision, audio, and language understanding in a single architecture. With 30 billion parameters but only 3 billion active per inference, the model delivers 9x higher throughput than comparable models while running on a single GPU. Early adopters include Foxconn, Palantir, and H Company, with Dell, Oracle, and Infosys evaluating it for production deployment.

Technology6 sources

Nvidia releases Nemotron 3 Nano Omni with 30B params for edge AI agents and multimodal tasks

Technology6 sources

Google donates Agent Payments Protocol to FIDO Alliance for secure agentic payments

Google has transferred ownership of its Agent Payments Protocol (AP2) to the FIDO Alliance, while Mastercard contributes its Verifiable Intent framework. The move aims to establish open standards for AI agents to execute autonomous transactions securely. AP2 v.0.2 introduces Human Not Present payments, enabling agents to make purchases based on pre-authorized user instructions.

Technology2 sources

Google donates Agent Payments Protocol to FIDO Alliance for secure agentic payments

Technology2 sources

Canonical reveals AI roadmap for Ubuntu: local inference and user control, not forced integration

Canonical has outlined its plan to integrate AI into Ubuntu Linux, taking a markedly different approach from Microsoft's Copilot strategy. Jon Seager, VP of engineering at Canonical, detailed how AI features will arrive as optional enhancements focused on accessibility improvements and local inference rather than cloud-dependent tools. Users retain full control with removable Snap packages instead of mandatory AI integration.

Technology9 sources

Canonical reveals AI roadmap for Ubuntu: local inference and user control, not forced integration

Technology9 sources

DeepSeek V4 closes gap with frontier models while slashing AI costs by 75%

Chinese AI company DeepSeek has released V4, its most powerful open-weight AI model yet, featuring 1.6 trillion parameters and a one million token context window. The model competes with leading systems from OpenAI, Google, and Anthropic while offering dramatic cost reductions—up to 75% off for developers. The release intensifies the Chinese AI price war and showcases architectural innovations that could reshape the global AI landscape.

Technology42 sources

DeepSeek V4 closes gap with frontier models while slashing AI costs by 75%

Technology42 sources

OpenAI releases Privacy Filter to scrub personal data before it reaches AI models

OpenAI launched Privacy Filter, a free open-source tool that detects and removes personally identifiable information from text before it reaches cloud servers. The 1.5-billion-parameter model runs on standard laptops, achieving 96% accuracy on industry benchmarks. Released under Apache 2.0 license, it addresses growing concerns about sensitive data exposure in AI workflows.

Technology2 sources

OpenAI releases Privacy Filter to scrub personal data before it reaches AI models

Technology2 sources

Former Meta exec launches nonprofit to help Gen Z find jobs as AI agents reshape entry-level market

Clara Shih, former AI executive at Meta and Salesforce, watched AI agents surpass her top employees and recognized a fundamental shift in the job market. Now she's launched the New Work Foundation to equip Gen Z with AI skills as they face the worst entry-level job market in 37 years, offering free tools to help young workers navigate an AI-dominated workforce.

Technology5 sources

Former Meta exec launches nonprofit to help Gen Z find jobs as AI agents reshape entry-level market

Technology5 sources

Anthropic Mythos discovers 271 software vulnerabilities in Firefox 150 before release

Mozilla's Firefox 150 includes fixes for 271 security flaws discovered by Anthropic Mythos Preview, a new AI model with advanced bug-finding capabilities. The AI identified vulnerabilities by analyzing unreleased source code, a task that would have required months of human effort. Firefox CTO Bobby Holley says defenders now have a decisive advantage in cybersecurity.

Technology20 sources

Anthropic Mythos discovers 271 software vulnerabilities in Firefox 150 before release

Technology20 sources

Moonshot AI's Kimi K2.6 coordinates 1,000 agent swarms running for days on complex tasks

Moonshot AI unveiled Kimi K2.6, an open-source AI model that orchestrates up to 1,000 collaborating agents to handle complex tasks autonomously. The model ran continuously for five days in one demonstration, managing monitoring and incident response without human intervention. But the release exposes critical gaps in enterprise orchestration frameworks not designed for such long-running, stateful execution.

Technology2 sources

Moonshot AI's Kimi K2.6 coordinates 1,000 agent swarms running for days on complex tasks

Technology2 sources

Manycore shares surge 140% in Hong Kong IPO as Chinese AI firm bets on spatial intelligence

Manycore Tech raised $160 million in an oversubscribed Hong Kong IPO, with shares rallying 140% on debut. The Hangzhou-based Chinese AI firm is pivoting from its real estate software roots to spatial intelligence, targeting robotics and world models. Co-founder Victor Huang, a former Nvidia engineer, believes robotics will define the era as the company leverages 500 million 3D assets to train AI for the physical world.

Business3 sources

Manycore shares surge 140% in Hong Kong IPO as Chinese AI firm bets on spatial intelligence

Business3 sources

Tech enthusiasts build local LLM servers on Raspberry Pi and phones, proving on-device AI works

Developers are successfully running local LLM servers on modest hardware like Raspberry Pi 5 and smartphones, eliminating subscription fees and privacy concerns. Google's Gemma 4 models have made on-device inference practical, with users reporting speeds of 5-8 tokens per second on single-board computers and phones. The shift challenges assumptions about the hardware requirements for self-hosted AI.

Technology7 sources

Tech enthusiasts build local LLM servers on Raspberry Pi and phones, proving on-device AI works

Technology7 sources

Anthropic's Model Context Protocol has critical flaw exposing 200,000 servers to remote takeover

Security researchers at OX Security discovered a critical architectural vulnerability in Anthropic's Model Context Protocol that enables remote code execution across 200,000 server instances and affects over 150 million downloads. The flaw is baked into MCP's official SDKs across Python, TypeScript, Java, and Rust. Despite repeated requests for a protocol-level fix, Anthropic declined to patch the issue, calling the behavior "expected."

Technology4 sources

Anthropic's Model Context Protocol has critical flaw exposing 200,000 servers to remote takeover

Technology4 sources

Mozilla launches Thunderbolt AI client targeting enterprises seeking data sovereignty

Mozilla unveiled Thunderbolt, an open-source enterprise AI client that lets organizations run self-hosted AI infrastructure without relying on cloud providers like OpenAI or Microsoft. Built on the Haystack framework, it promises data sovereignty, end-to-end encryption, and freedom from vendor lock-in while supporting multiple AI models and platforms.

Technology5 sources

Mozilla launches Thunderbolt AI client targeting enterprises seeking data sovereignty

Technology5 sources

Tiny Corp enables Nvidia GPUs on Macs with Apple-approved driver for AI workloads

Tiny Corp has developed TinyGPU, an open-source driver that brings Nvidia GPU support back to macOS after Apple dropped it in 2018. The Apple approved driver allows external GPUs like the RTX 5090 to connect via Thunderbolt or USB4 without hacks. While performance lags behind native Metal, the driver opens new possibilities for AI experiments on Mac.

Technology2 sources

Tiny Corp enables Nvidia GPUs on Macs with Apple-approved driver for AI workloads

Technology2 sources

Salesforce unveils Headless 360, turning its entire platform into infrastructure for AI agents

Salesforce introduced Headless 360 at its TDX developer conference, exposing every capability in its platform as APIs, MCP servers, and CLI commands for AI agents. The initiative represents a fundamental operating model transformation, making the CRM giant's entire ecosystem accessible to coding agents without requiring a graphical interface. With over 100 new tools and skills, the company is betting that most future code will be written by agents rather than humans.

Technology9 sources

Salesforce unveils Headless 360, turning its entire platform into infrastructure for AI agents

Technology9 sources

Nvidia releases Ising open AI models to solve quantum computing's biggest bottlenecks

Nvidia unveiled Ising, a family of open source AI models designed to tackle quantum computing's critical challenges in calibration and error correction. The models deliver 2.5x faster and 3x more accurate performance than existing tools, while sparking a rally in quantum computing stocks as IonQ surged 50% and major research institutions rushed to adopt the technology.

Technology18 sources

Nvidia releases Ising open AI models to solve quantum computing's biggest bottlenecks

Technology18 sources

Prev 1…131415…28 Next