gpt-oss-120b only has 5B active parameters, so its not surprising Qwen3.8 27B outperforms it (Qwen3.8 is also ~13 months newer, which is forever in LLMs)
And Qwen3.8-27b is still better at coding than opus 4.1.
Yes, if you list off models 27b is better than it’s all older models. But that’s my point - newer models are better than older models at the same AND much smaller size. That’s because model size matters less than they say. Training data and model architecture matter more.