Outperform in what way? Reasoning models may be able to solve problems correctly a bigger percentage of time, but they burn many tokens to get there. So they’re much less efficient, both in latency and ultimately environmental cost.