Price/performance is as advertised in my limited testing. I happened to be testing the same task with Haiku, GPT-mini, and Deepseek v4 0731 yesterday, so I had a reasonable baseline to compare the model to.