Tony Huang 评 Qwen4-Max:面向科学发现,SWE 提升有限
Interesting thing here. Qwen-4 appears to be built w/ more science discovery in mind. It seems like SWE is not its strong suit based on these benchmar...
AI 导读
Tony Huang 评价 Qwen 发布的 Qwen4-Max,认为其更像面向科学发现而设计,评测显示 SWE 并非强项,相比上一代 Qwen 提升不大。Qwen4-Max 基于 MoE 架构,具备原生视觉能力和 1M-token 上下文窗口,官方称相比 Qwen3.8-Max 少用最多约 50% token,且单任务平均成本下降。