Open-Source AI
Open models — Llama, Mistral, DeepSeek, Qwen and others — are catching up to closed labs and let you run AI on your own hardware, vendor-free. That shifts the balance of power, the safety debate and the economics of the field. Here is everything on open-source AI: open-weight releases, licenses, local deployment and the race against closed models.
JetBrains released Mellum2 — an open 12B MoE model for fast tasks in AI pipelines
JetBrains open-sourced Mellum2 under Apache 2.0 — a compact MoE model with 12 billion parameters, trained on 10.6 trillion tokens for integration into multi-model…
NVIDIA JetPack 7.2: AI agents on edge devices with a single deployment command
NVIDIA released JetPack 7.2 for the Jetson platform — a single CLI command deploys a full agent stack with privacy and memory optimization.
Meta explains the difference between GPU, CPU, and its own MTIA chips in AI infrastructure
Meta launched an educational series and explains how GPU, CPU, and custom MTIA chips provide the computing power for its AI systems for billions of users. *Meta is…
Liquid AI released LFM2.5-230M: 213 tokens/s on Galaxy S25 and support for llama.cpp
graphlens-mcp builds a code graph with one command and cuts token usage by 10–23x
OpenAI unveils GPT-5.6: three models, Sol, Terra and Luna, for different tasks and budgets
Google backed AI regulation — but only in a form that suits it
Macron and Modi warn G7: the US could shut off AI at any moment
At the G7 summit, Macron and Modi said the US is capable of cutting off access to American AI tools without warning — and a recent Anthropic outage turned that…
Mistral AI Gives Administrators Fine-Grained Control Over Enterprise Connectors
Mistral AI launched six new features for enterprise connectors: tool-level access management, multi-account support, and an MCP connector debugger.
ML Red Teaming for LLMs: From Hallucinations to Data Leaks — Testing in Practice
Z.ai releases GLM-5.2: real million tokens and two levels of deep thinking
London leads Europe's AI race: Bloomberg breaks down the reasons behind the British tech boom
Anthropic Launches Claude Corps: $150M and 1,000 AI Specialists for US Nonprofits
How to Build an Agent Workspace on QwenPaw with Custom Skills and Streaming API
Step-by-step tutorial for deploying QwenPaw — a framework for creating AI agents with support for multiple model providers, custom skills, and streaming API in…
A Single Suffix Breaks Any LLM: Researchers Found One Universal Refusal Vector
Different adversarial attacks on LLMs — GCG and AutoDAN — turned out to be two paths to the same place: a single refusal vector that controls the entire model's…