Qwen3.8 27B - Tried A Real Life Test

# מהמקור

I was hearing a lot about Qwen3.8 27b so I wanted to give it a shot. I ran it on my system (Mac Studio, M2 Ultra 24-core CPU, 76-core GPU, 128gb of ram) with opencode / ollama. Generation speed by quant: MLX 4-bit: 51.5 tok/s (ran the tasks on this one) Q4_K_M: 33.4 tok/s Q8_0: 21.3 tok/s Quality on well-specified tasks: CRUD bank task: passed SemVer 2.0 comparator: 26/26 Recursive-descent calculator (right-associative): 24/24 Latency observations: Calculator task: ~11 min to complete One prior attempt hung ~6.5 min and produced zero files before retry Failure mode: Ships confident-but-wrong co