Have you tried qwen 3.8 flash next? So far I've tested it only briefly but it's definitely a lot better compared to other local llms that I can run (~30B class models). For reference I am on 16 vram and 32 ram, older ddr4 platform.