AI Intel

GLM-5.3-Flash (ox-alpha) can now be run locally using Unsloth GGUF, enabling 3-bit quantization on 128GB RAM, and reportedly rivals Claude Opus 4.8 o…

@Kimi_Moonshot's Kimi K3, an open-weight model, is now generally available and integrated into GitHub Copilot, providing frontier-level agentic codin…

Gemini 2.5 Pro and Gemini 2.5 Flash models are now Generally Available (GA) and utilized across all Gemini Code Assist user tiers, which also receive…

A new arXiv paper introduces CP-Agent, an agent specifically designed to streamline crystal plasticity simulation workflows by reducing manual config…

A Hacker News discussion is buzzing with questions regarding whether Claude Opus 5.5's performance has been 'nerfed' or reduced.

AMD's new 256-core EPYC processor features 16-channel DDR5-12800 memory, achieving 91% of an RTX 5090's memory bandwidth, sparking interest in the Lo…

A Reddit post suggests that the era of subsidized compute for ChatGPT Pro is ending, with the old $200 plan's limits being halved and a new $500 plan…

The Deepseek Harness app has been released, generating excitement and downloads within the LocalLLaMA community on Reddit.

A new arXiv paper presents SMARtCARE, a privacy-preserving, bounded-autonomy agentic AI system designed for clinical decision support.

A new arXiv paper introduces 'Witeness Overlap,' a method for auditing the directional provenance and relationships within open-weight model families.