Early this morning Beijing time, Opus 5.2 quietly initiated grayscale testing in Claude Code. Developers noticed significant differences in its performance compared to the web-based Opus 5 during invocation. Through packet capture analysis, it was discovered that the backend model identifier has been updated to Opus 5.2. Additionally, an internal risk report from Anthropic revealed that the company is activating its yet-to-be-released 'nuclear-grade' Model 2, which has already taken over most of the company's code-writing tasks and is expected to replace 85% of the research team's workload.
