Anthropic has initiated a gray-box test of the Claude Opus 5.2 model within the Claude Code environment. Developers identified the update by inspecting request statuses, noting that the system is routing tasks to the new version despite the interface label remaining unchanged.
Users report that Opus 5.2 demonstrates increased generation speed and improved performance in code generation and long-text processing compared to the previous iteration. The model is capable of autonomous iterative refinement, completing complex tasks without requiring manual prompts or follow-up instructions.
Internal reports suggest that Anthropic is also developing a highly advanced model codenamed Model 2. This model reportedly achieved a score of 62.8% on the CoBench v2 benchmark, outperforming the Mythos 5 model by 12.5 points. It is currently being used internally to handle a significant portion of the company's code generation tasks.
According to internal documentation, Anthropic aims to utilize its advanced AI models to replace up to 85% of its research team's workload. The company is preparing three distinct strategies to compete with OpenAI's upcoming GPT-6 Astra, with a full market release of the new model family expected between the end of this month and the end of next month.
