按显存档位本地跑 agent 建世界：DeepSeek V4.1 Flash 出现在 256GB/512GB 配置
作者按 16GB 到 512GB 显存/内存分档，用本地 agent 各建一个世界；DeepSeek-V4.1-Flash 出现在 2×Spark 256GB、Mac 256GB 和 4×Spark 512GB 三档配置里。
That's it for today. I let local AI agents build one world per memory class. Today we saw: 16 GB: Qwen3.8-27B on an RTX 5060 Ti, @MiaAI_lab's kit https://niklaslenz.com/v2/qwen38-27b-exl3-3-0bpw-mtp-5060ti/ 32 GB: Qwen3.8-27B on an RTX 5090, @MiaAI_lab's kit https://niklaslenz.com/v2/qwen38-27b-exl3-4-0bpw-mtp-5090/ 1 SPARK 128 GB: Qwen3.8-Flash-Next, @MiaAI_lab's recipe on @ashxhart's TensorFold https://niklaslenz.com/v2/qwen38-flash-next-1spark-tensorfold-fp4/ 2 SPARKS 256 GB: DeepSeek-V4.1-Flash, @MiaAI_lab's recipe on TensorFold https://niklaslenz.com/v2/deepseek-v41-flash-2sparks-mia-tensorfold/ MAC 256 GB: Qwen3.8-Flash-Next on oMLX by @jundotkim https://niklaslenz.com/v2/qwen38-flash-next-mac-omlx-mtp/ MAC 256 GB: DeepSeek-V4.1-Flash, gnahzhang's recipe on oMLX https://niklaslenz.com/v2/deepseek-v41-flash-mac-lsq3-g128-omlx/ 4 SPARKS 512 GB: DeepSeek-V4.1-Flash, @MiaAI_lab's TP4 recipe https://niklaslenz.com/v2/deepseek-v41-flash-4sparks-mia-tp4/ Models by @Alibaba_Qwen and @deepseek_ai. You can play every world in your browser on its page.