Open-weight & owned stack · Quantization efficiency
Halogen Improves Qwen3.8 Long-Context Local Serving
Halogen 0.12.0 reportedly increased Qwen3.8-Flash-Next decoding to 38.3 tokens per second at roughly one-million-token context on Strix Halo.
Read the original at reddit.comOpens the publisher's site in a new tab