Is Gemma 4 E4B still the best small model for a local, tool-calling personal assistant? What am I missing?

I run a local assistant on a Mac (MLX via oMLX, 24 GB M5 Pro). It needs to call tools reliably across mail, calendar, notes and files, chain a few calls, and reply in a short, calm voice. It also has to synthesize/extract important information from my messages, emails, and notes. It also has to be able to read long emails and docs, use my voice to ask questions and take screenshots of my screen. Gemma 4 E4B has been the best model I’ve tried so far since it supports text, image, and audio. However, it is quite old now(~4 months old) and I wanted to know if there has been any new small models(under 8B or 6GB in size) that outperform Gemma 4 E4B in tool calling, chaining tools, long synthesis, fact extraction, and in size while having multimodality features. I have tried ling 3.0 tiny, the LFM model family, and qwen 3.5 4B/9B(low quant) but have had no success so far. Is there anything I'm missing? Is Gemma 4 E4B the best model for my usecase?