2026-11-02
AlphaLlama v2 decodes Qwen3.8-Flash-Next at 602 tokens a second on one RTX 5090: the numbers, how we measured them, and where it is and isn't faster.