CommandCode Rebuilds Its Read Tool to Save Billions of Tokens
Ahmad Awais announces an updated read tool benchmark for CommandCode AI and explains how the rebuilt read_file was engineered for open models, claiming billions…
Ahmad Awais announces an updated read tool benchmark for CommandCode AI and explains how the rebuilt read_file was engineered for open models, claiming billions…
CommandCodeAI v1.47.0 now natively supports most file formats, with the maintainer saying it is faster and more token-efficient than custom skills. Update via `…
Qwen introduces E-Commerce Bench, a new benchmark where AI agents run virtual online stores for 365 days with ¥100,000, testing long-horizon business reasoning.
Qwen's new E-Commerce Bench tests AI agents on year-long autonomous online store management with ¥100,000 starting capital, real-market data, and hundreds of su…
OpenAI is committing $1 billion to subsidize Daybreak access and frontier AI capabilities for frontline defenders of essential services and critical infrastruct…
Sam Altman highlights GPT-6 Astra's potential for scientific discovery, pointing to an OpenAI repository where the model formalized in Lean a proof that infinit…
OpenAI CEO Sam Altman apologizes for the messy Astra rollout, promises compensation for paid users, and says broad access for API customers and ChatGPT subscrib…
OpenAI says paid ChatGPT users without Astra access will receive one banked reset for each day they are locked out, starting today. The first reset is expected…
Ollama's new interactive menu lets users discover and launch models, agents, and coding tools without memorizing commands or searching shell history, making loc…
MiniMax's official account retweets a live H3-powered demo: an infinite, point-and-click adventure where every picture branches into a new world.