AI Intel

Surge AI conducted a blind human evaluation for Mistral Large 4, where Le Chonk ranked #1 in coding quality among open-weight models.

Mistral AI announced a partnership with Mozilla to enhance privacy, control, and choice for users leveraging AI for online browsing.

Merve has released a local AI slide deck covering topics from prefill vs. decode, MoE vs. dense models, VRAM vs. unified memory, quantization, and sp…

Hugging Face retweeted the release of Liquid AI's Open d1 family of open-weight multimodal decision models, including d1-3B (text+vision) and d1-omni…

Gemini Code Assist has rolled out multiple updates, including the General Availability of Gemini 2.5 Pro and 2.5 Flash models, Agent mode in VS Code…

A new arXiv paper introduces a cognitive architecture for Adaptive Workflow Intelligence, aiming to address the brittleness of AI solutions in enterp…

A new arXiv paper presents KVFetch, a method that employs temporal prefetching to optimize KV cache compression for efficient LLM inference.

The Saluki 27B model claims to achieve 96% of Qwen 3.8's performance while being approximately one-seventh of its size.

Researchers successfully removed prior political alignment from the Qwen3.6-35B-A3B model, reducing censored/propagandistic responses from 89.8% to 2…

PodiumHQ is alpha testing OpenAI's new Decisions API to enable their agents to select appropriate workflows more quickly and efficiently.