Comparing to Opus is a little unfair, a comparison against Haiku would be more fair. And for a really fast cloud model, I'd be interested to see how latency stacks up against GPT-5.3 Instant or Gemini 3.1 Flash Lite.