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.
OpenMythos: Open-source PyTorch reconstruction of Claude Mythos architecture with 770M parameters
Developer Kye Gomez reconstructed the presumed Claude Mythos architecture from Anthropic from scratch — without leaks, based only on public research.
LangChain Loses Reasoning Content in CoT Models: How to Fix the LLM Provider Bug
LangChain chat classes fail to preserve the reasoning block when working with Chain-of-Thought models—users wait silently while developers are forced to fix it…
Amazon to Invest Another $25 Billion in Anthropic and Reserve 5 GW of Capacity
Amazon has agreed to new $25 billion investments in the creators of Claude. Anthropic has reserved up to 5 GW of AWS computing capacity for service scaling.
Hugging Face: open-source AI gives defenders the same capabilities as attackers
OpenAI launched Chronicle in Codex — a Windows Recall equivalent for developers
SpeShu.AI launches feature crowdsourcing: Habr users decide what goes into the next patch
Claude Code barred to minors, Qwen Code goes paid, and Opus 4.7 uses 30% more tokens
Amazon SageMaker and DVC: end-to-end traceability of ML models from data to prediction
AWS showed how to combine DVC and SageMaker MLflow Apps for full traceability of the ML pipeline — at the dataset and individual-record levels.
China’s OpenClaw frenzy turns into a gold rush for AI companies
The open-source AI agent OpenClaw has sparked mass demand in China: users are renting servers and buying subscriptions just to test it, bringing unexpected profits…
Rana el-Kaliouby: "boys' club" in AI threatens to entrench the gender income gap
Mistral Forge: training enterprise AI from scratch on proprietary data
Multiverse Computing launched a public API with compressed OpenAI, Meta and DeepSeek models
Cursor admits: its new coding model is built on Chinese Kimi from Moonshot AI
Zuckerberg launched Meta initiative to support entrepreneurs and AI adoption
Meta is launching a new program to support entrepreneurs: the company wants to help small businesses actively use AI for growth, customer acquisition, and scaling…
Google unveils TurboQuant — an algorithm that compresses AI working memory sixfold
Google has developed TurboQuant, an algorithm that theoretically compresses neural networks’ working memory sixfold — the internet is already comparing it to Pied…