After so much testing with different prompts and models, ended up wording a good query for decompiling with r2ai/decai.
The first screenshot shows the result for: Claude3.5, Gpt4o and Qwen2.5 (local) for a password checking function in Swift.
The second one is from r2ghidra, but GHIDRA/IDA/BN results are at the same level of uselessness


