3 ms·

I'm using Qwen3Next on my MBP. It uses around 42GB of memory and, according to Aider benchmarks, has similar perf to GPT-4.1 https://huggingface.co/mlx-communi

• binary132 1y ago
  Just waiting on llama.cpp support :) I usually use GPT-oss-120B with CPU MoE offloading. It writes at about 10tps, which is useful enough for the limited things I use it for. But I’m curious how Q3 Next will work (or whether I’ll be able to offload and run it with GPU acceleration at all.) (4090)