On September 3, OpenAI announced its next-generation AI model, "GPT-6 Astra." The model is characterized by enhancements in computer operation, coding, research, and complex multi-step tasks. According to the company, it recorded a score of 72.6% (including partial credit) in evaluations of offline target tasks in OSWorld 2.0.
Prior to this announcement, Anthropic released "Claude Fable 5.1" on September 1. This model is designed for workflows involving the use of multiple applications over several hours, progressing through tasks by planning, utilizing tools, and recovering from errors.
Additionally, Google announced "Gemini 3.8 Flash" on September 2. According to Google, the model significantly enhances software development, agent-based tasks, and multi-step inference capabilities; this marks the third update for the Flash series in just six weeks.
With the emergence of major models from these three companies within a few days, the axis of AI competition is shifting beyond simple response accuracy toward "completion power"—the ability to execute a business process from start to finish, including browser and application manipulation.
Source: