OpenAI’s GPT-6.2 models improve some benchmarks while coding results split
OpenAI released GPT-6.2 Astra and Sol with lower Astra API rates and mixed independent coding results.
OpenAI’s GPT-6.2 safety tests expose a gap between AI monitors
OpenAI’s system card reports sharply different detection results across monitors in simulated GPT-6.2 security tests.
Microsoft rolls out a portfolio of specialized MAI models
Microsoft announced a family of specialized coding, security, image, audio and decision models across its existing products.
ONNX Runtime’s GitHub repository became unavailable while packages stayed online
The canonical ONNX Runtime GitHub repository returned 404 responses while existing packages remained listed in major registries.
Coding agents sped up pull requests as review delays grew in a large firm study
A study of 655 firms linked coding-agent adoption to more pull requests and longer elapsed review times.
A search experiment finds fewer publisher clicks in forced AI Mode
A field experiment found that forced Google AI Mode reduced click-through to publishers under its study definition.
Nvidia’s revenue geography does not show where its chips end up
Nvidia’s latest revenue table classifies sales by direct-customer headquarters, not final chip destination.
Tencent’s Hy4 Flash puts a large checkpoint behind a conditional license
Tencent’s Hy4 Flash offers downloadable weights, but its full-precision repository is about 440.5 GB and its license has commercial conditions.