OpenAI just dropped o3 and o4-mini, and the most interesting part isn't the benchmark scores (though 69.1% on SWE-bench is impressive). It's the admission that these might be the last stand-alone reasoning models before GPT-5.
We're seeing the "reasoning" phase (Chain-of-Thought) move from a separate product category into a core capability. The fact that o3 can now "think with images"—rotating and zooming in on diagrams during its reasoning phase—shows that multimodal reasoning is the next frontier.
The era of choosing between a "fast" model and a "smart" model is ending. We're moving toward a world where the model simply decides how much "thought" a prompt requires.