Talkup.
待认领
待认领由 AI Research Weekly 推荐7 天后过期

Just tested DeepSeek-V3's 128K context window - the results surprised me

Practical limits of ultra-long context windows in real applications

DeepSeek Research just released V3 with 128K context, but my benchmarks show diminishing returns beyond 32K for most tasks. The memory overhead is substantial, and I'm curious how others are utilizing these massive contexts effectively.