Researchers led by the Allen Institute for AI describe 'byteification', a two-stage method that retrofits standard language models to work on raw bytes instead of word-piece tokens. The converted models came close to their originals and beat earlier byte-level models of similar size.
OpenAI began rolling out a GPT-6 model in ChatGPT that can answer with interactive interfaces. Paid tiers got it first, with free users following the next day.
Abu Dhabi's Technology Innovation Institute released Falcon-ASR, a 1.6B-parameter speech recognition model. It reports a 20.92% average word error rate across six Arabic test sets.
NVIDIA reports that specialized versions of Nemotron 3 scored above the gold thresholds at both the 2026 International Olympiad in Informatics and the International Mathematical Olympiad. The runs were unofficial.
The FCC authorized SpaceX to deploy a 15,000-satellite constellation for direct-to-device service, while deferring some requests. SpaceX must launch half by October 2032 and the rest by October 2035.
Mistral opened a public preview of Mistral Large 4, a natively multimodal model with about 1 trillion total and 52 billion active parameters. The French lab says the weights will be released by the end of October.
The UK government said it accepts every recommendation from the National Commission on regulating AI in healthcare. Medicines regulator MHRA plans a consultation on how software and AI devices are classified by spring 2027.
Google DeepMind released EmbeddingGemma 2, a 740-million-parameter open model that maps text, code, images, audio and video into one shared embedding space. It is released under the Apache 2.0 license.
Hugging Face's post-training team released an open framework and guide for reinforcement-learning a model inside unmodified coding agents, built on OpenEnv, the Harbor task framework and the TRL trainer.
The llama.cpp server now supports decision models through a new endpoint. These models read an input once and return a probability for each option you give them.