BLAST RADIUS

Open-weight & owned stack · Quantization efficiency

Halogen Improves Qwen3.8 Long-Context Local Serving

Sep 19, 2026

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

More in Open-weight & owned stack