Is there a better small model than Qwen3.5 4B for a fast local AI assistant?

I'm currently using Qwen3.5 4B as the brain of my local AI assistant because my hardware is relatively limited. One thing I really like about it is the speed. On my system, I'm getting around 40–50 tokens/sec, which makes the interaction feel surprisingly close to real-time. So I don't want to move to a much larger model if that means losing that responsiveness. What I'm wondering is whether there are any newer or alternative models in roughly the same size/memory/compute range that are noticeably better than Qwen3.5 4B for my particular use case. My priorities are mainly: - General-purpose conversation and communication - Natural responses and understanding - Reasoning and following instructions - Multilingual conversations and translation - Context understanding - Tool/function calling mainly for triggering actions in my assistant, rather than writing code - Reliability when interacting with tools - Ideally good performance while keeping the model fast and lightweight Essentially, I'm trying to build a local assistant where the model acts as the reasoning/conversation layer, while other tools handle things like actions, APIs, automation, etc. So my question is: If I'm already using Qwen3.5 4B and getting ~40 to 50 TPS, is there anything in a similar practical memory/compute footprint that you'd personally consider a significant upgrade for general conversation + multilingual ability + tool calling? It doesn't necessarily have to have fewer parameters. I'm more interested in overall capability relative to the amount of RAM/VRAM and compute it requires. I'm happy with Qwen3.5 4B, so I'm not looking for something that's only marginally different. I'm wondering if there's something that makes you think