Istari won a Department of War prototype other transaction agreement worth up to $50 million to build the Mission Engineering Automation Testbed.
GPT-5.4 is also more reliable, producing 18% fewer errors and 33% fewer false claims than GPT-5.2, according to OpenAI.
OpenAI launches Codex for Windows, letting developers run multiple AI coding agents, automate testing tasks, and sync projects seamlessly across Mac and Windows.