Anyone tried Nanbeige4.3-33B?

Nanbeige4.3-33B , a 33B parameter model with a 512K context window, and capabilities in reasoning, coding, and agentic tasks. Has anyone tested it? Benchmarks are insane but how does it compare in real world use cases? I understand that it is a kv cache hog. But how does it perform? Would love to hear your experiences!