local llm 发展得如此地迅速，以至于让所有手写的文章都瞬间过时。 跑高智能去审查版本 llm 的门槛已经非常低了，最低三千美元起，丰俭由人。 把 qwen 3.8 flash next thinking effort 调高，并不比其他模型差太多。 1* dgx spark / 3090+ 128gb ddr4: qwen 3.8 flash next 2* dgx spark：glm 5.3 flash 4* dgx spark：deepseek 4.1 flash 1017 https://artificialanalysis.ai/?models=step-5%2Cinkling%2Cglm-5-3-flash%2Cmimo-v2-6-pro%2Cgpt-6-astra%2Cclaude-opus-5-5%2Cmistral-large-4%2Cmuse-spark-open%2Cclaude-fable-5-1%2Cgrok-4-7%2Ckimi-k3-1-pro%2Creflection-beam%2Cgpt-6-luna%2Cclaude-haiku-5-5%2Cgemini-3-8-flash%2Cqwen3-8-27b%2Cqwen3-8-max%2Cclaude-sonnet-5-5%2Cgpt-6-1-sol%2Ck2-horizon-375b-a23b%2Cglm-5-3%2Cgemini-4-argon%2Cminimax-m3-1-flash%2Cnemotron-3-5-super-vl%2Cmuse-wonder-full%2Cdeepseek-v4-1-flash-1017%2Cmuse-wonder%2Cqwen3-8-flash-next