Microsoft's AI research group has detailed MAI-Base-2, which it says is significantly more efficient than leading open-source alternatives. It is a base model rather than a finished assistant, meant to be turned into a full LLM through further training.
Elorian has released Glimpse, its first AI model, built to reason visually. According to the company's own evaluations, it beats GPT-6 Astra on scientific chart reasoning and short video captioning, with competitive results on other visual tasks.
Data from Vercel's AI Gateway, a unified API service similar to OpenRouter, shows Jev has been adopted faster than any other model on the platform.
Gemini 4 Argon scored 97.5 percent on ARC-AGI-3, making Google the second lab after OpenAI to pass 95 percent. Yet the result falls to 49.8 percent in the standard harness, and the model still trails rivals in maths and software engineering.
Qualcomm CEO Cristiano Amon said on Alex Heath's Sources podcast that leading AI companies are asking for phones able to run models of at least 100 billion parameters by 2028. The remark signals how far on-device AI ambitions are stretching.
Since the summer, Mistral has been serving Zhipu AI's GLM model on its platform, a move that drew controversy. The JDN argues it was a financial and technical bet that paved the way for Mistral Large 4, its new trillion-parameter model.