Models
6 stories in this blend
Superintelligence·Sep 29OpenAI prepares multi product release event alongside new model previews
OpenAI chief executive Sam Altman teased upcoming announcements ahead of the company's developer event. Reports suggest the company will launch several software tools alongside initial previews of specialized security systems.
The Neuron·Sep 15Reward AI introduces OM 1 control model for robotic hardware
Reward AI launched OM 1, a robotics foundation system trained on human movement data. The software works across multiple hardware configurations, including robotic arms and humanoid machines.
Superintelligence·Sep 4OpenAI releases GPT-6 Astra with focus on autonomous task alignment
OpenAI launched GPT-6 Astra, introducing a flagship model built to perform lengthy computer workflows without stepping outside safety parameters. On alignment tests, the model consistently refrained from bypassing review filters or attempting unauthorized actions. The underlying system relies on hardware clusters in Texas and new context retention mechanisms designed for multi step technical work.
Superintelligence·Sep 4Google DeepMind upgrades weather model to deliver street level forecasts
Google DeepMind released WeatherNext 3, a weather forecasting model operating at a five kilometer resolution with hourly updates. The system processes direct geostationary satellite feeds to reduce data latency and improve precipitation accuracy by up to sixty percent. It is being integrated directly into Google Search, Maps, and the Gemini mobile application.
AI Daily Brief·Aug 26AI developers reduce prices on flagship agents and models
Access to GrokBot is moving to lower subscription tiers, including Cursor Pro and SuperGrok, making it significantly cheaper for developers to use. Meanwhile, OpenAI cut API rates for its GPT-5.6 Sol model. The price adjustments reflect growing pressure across the industry as enterprises adopt tiered model strategies.
Superintelligence·Aug 21Unidentified AI model achieves top marks on programming evaluations
An undisclosed model named Ox Alpha recorded leading benchmark results on programming tasks through the OpenCode platform. The system supports multi-modal inputs and expanded memory capacity, though its creators have not disclosed their identities.
