Alibaba's Qwen4-Max is a 2.4T-parameter agent model that cuts per-task costs by about 60%. It went live on You.com's Inference Router on day one, but its weights stay closed.
Google Research open-sourced StreetReaderAI, a multimodal map prototype that lets blind and low-vision users navigate Street View with voice, keyboard, and live AI scene descriptions.
A multi-agent reinforcement learning system from DeepReinforce swept three consecutive live Codeforces rounds, outperforming every human including legendary grandmasters.
NVIDIA's LidarVerse packs generation, forecasting, and resimulation of LiDAR scans into one real-time world model that runs at 10 FPS on a single H100.
OpenAI's Economic Research team analyzed 580,000 messages from 1,819 scientists in a pilot program, finding researchers finish work nearly twice as fast.
Artificial Analysis launched a Time to First Audio benchmark for text to speech models, showing streaming systems answer in under 400 ms while non-streaming peers take 2 seconds.
A new paper argues that self-attention works like the Dyson equation from many-body physics, which would explain why sum-pooling models break down on correlated quantum states.
OpenAI hands Apollo, METR, Transluce, and Goodfire deep access to internal models, training runs, and interpretability tooling to stress-test alignment.