登録して招待リンクを共有すると、動画再生報酬と紹介報酬を獲得できます。

Yum⋆₊˚
@yuhasbeentaken
Growth enginee|prev @Stanford @tiktok_us @Weixin_WeChat|building @Offloop best workspace for teams
参加 July 2023
1K フォロー中    6.7K ファン
Kimi K3 doesn’t make sense for agentic coding right now. a few things i’ve noticed: 1. Claude at $20 can go further than Kimi at $200. that pricing difference is hard to justify when the actual amount of useful agentic work can be lower. 2. the $200 Kimi limits can disappear ridiculously fast. long coding sessions and multi-step agent runs burn through the allowance way too quickly. 3. planning and Max reasoning make it even worse. K3.1 Pro is good at planning, but using it heavily for orchestration can destroy the quota in just a few serious sessions. 4. Kimi has also become noticeably slower. even when i still have usage left, i’d often rather send the execution work to DeepSeek or Claude than wait on Kimi. 5. the weird part is that Kimi is actually good at planning and code review. the problem is the economics. it makes much more sense to use it selectively for architecture/review and let a cheaper model do the repetitive execution. for agentic coding, i’d rather use: > Claude/Opus for planning and review. > DeepSeek V4.1 Flash for execution.
もっと見る