Similarly, many of my informal tests have started passing with Gemini 2.5 that never worked before, which makes the 2025 era of AI models feel like a step change to me.