I keep a close eye on Microsoft's Phi models and test them out for my own projects, because I really believe on-device AI is the next big thing.
The new Phi-4 Reasoning models (mini, reasoning, and reasoning-plus) are now out. For small models, their reasoning ability is genuinely top-tier right now.
That combination of efficiency and real reasoning power is why the Phi-4 family (maybe alongside models like Gemma) is so promising for actually getting powerful AI running locally on our devices soon.