Qwen3.5 reports

generation:

11.0 tokens/s

quant:

Q4 (GGUF)

Reported by the source; GPU count, offloading and concurrent requests can change these figures. Prompt-processing speed is shown only when the source reports it; generation speed cannot tell us time to first token. Check the full setup before comparing.

User reports Qwen 3.5 9B Q4 with MTP at 11 t/s output at 70k context on a pair of AMD FirePro D700 GPUs in an old Mac Pro. Setup is llama.cpp on Vulkan across the two D700s with 12gb of VRAM, in a machine with 64gb of RAM. Qwen 2.5 coder q4 runs at 22 t/s output at 70k context on the same machine. The user finds Qwen 3.5's planning output significantly better than Claude Sonnet 4.6 on a complex C# .NET 10 app.

quant:

WinterMix58 (MLX)

User reports a new MLX quantization method, WinterMix, for Qwen3.5-122B-A10B, with two builds: WinterMix58 at 82 GiB and WinterMix48 at 68 GiB. Benchmarks show perplexity improvements over existing MLX quants. The user is enthusiastic about the results and the method's advantages for agentic workflows on Apple Silicon.

User reports a model based on Qwen3.5-4B, trained on 8xH100 for 3 days. It supports Safetensors, GGUF and MLX weights, requires as little as 4GB VRAM, and is available in GPTQ, W8A8, FP8, Q4 and Q6 quantizations. It has been tested with vLLM, SGLang and llama.cpp.