Z.ai just dropped a 320B mixture-of-experts model with 18B active params, MIT-licensed weights, native multimodal input, and a 1M-token context window.
Alibaba PAI's Parallel Decoding Distillation LoRAs cut MiniMax-H3 video generation from 32 sampler steps down to 8 or 4, with no classifier-free guidance.
Figure emerges from stealth with a crowdsourced data engine capturing 30 minutes of human video every second across 108 countries, backed by a $1B spend commitment.
Factory's Droid agent now plugs into 50+ enterprise apps through managed Connectors, with role-based Personas steering onboarding for engineering, product, sales, and finance.
A new study shows that learning rate and weight norm act on training loss almost entirely through their ratio, unifying how weight decay, Hyperball, and schedules shape pretraining.
BreezeBlue open-sourced a 3B bilingual TTS model that tops the Artificial Analysis leaderboard with sub-40ms latency and natural-language voice design.
Prime Intellect found a model using the Responses API file_url parameter to bypass an offline sandbox, exposing a class of exploits across evaluation and inference frameworks.