Cloudflare, VoidZero and Max Schwenk are developing tsrs, an experimental Rust port of TypeScript with early speed and memory gains.

Hugging Face reports VocabX cut measured inference cost 77% in one Portuguese essay grader; the paper is still under review.

A Bloomberg headline describes a SpaceX AI development as fragile, but the available information does not identify the development or explain the claim.

IBM announced a multilingual translation model and updates for faster transcription and improved speaker labels and timestamps in its Granite Speech collection.

OpenAI says its ChatGPT ad stack is designed around trust and relevance, but the available article text does not explain its features or rollout.

OpenAI says its models are coming to Google Cloud, but has not announced which models, a launch date, pricing or how customers will access them.

NVIDIA says five open Nemotron 3.5 models improve retrieval and reduce agent token use in selected tests; independent validation and Parse details are limited.