Claude opus 5.2 launches with enhanced speed and reduced laziness

On September 15, 2026, Anthropic quietly released Claude Opus 5.2, which began a gray-scale test in Claude Code. Developers noted a significant performance shift when using Opus 5 in the code environment, indicating a potential skip over the 5.1 version. The update is characterized by three key improvements: faster response times, increased accuracy, and a more proactive approach to task completion. Developers reported that Opus 5.2 does not require prompts to continue complex tasks, instead engaging in continuous self-iteration until completion. A method to identify if one is in the gray-scale test involves asking about 'Tibo the Resetter,' a reference absent from older training data. Those using Opus 5.2 could accurately respond, distinguishing it from the web version. Anthropic’s internal reports suggest Opus 5.2 is part of a broader strategy against OpenAI’s GPT-6 Astra, with plans to launch the 5.2 family by year-end. The company also has two other models, including Model 2, which outperformed existing models on CoBench v2, and RSI, which could replace up to 85% of research team work.