中国小企业「猫鼠游戏」用 Opus 给客户设计 PCB,作者质疑编码榜单可信度
xiaosun86 · x · 2026-11-11
A substantive observation: coding benchmarks mask how imbalanced Opus is versus other models — Chinese small businesses reportedly play a cat-and-mouse game just to keep using Opus for real customer work like PCB design.
The author claims pipelines collapse immediately when switching to alternatives like Argon or Astra — "literally tokens in, revenue out." Conclusion: those benchmarks are no better than Jim Cramer's stock picks, with real revenue dependence badly diverging from leaderboard rankings.
所属事件:中国中小企业为用 Opus 与封禁玩猫鼠游戏(3 条相关)→
「模型」频道最新
- 曝 ChatGPT 测试人脸与声纹验证，用于确认账号变更 — nima_owji · 2026-11-11
- Anthropic 喊话放慢 AI,转头就发了 Fable 5.2 — haider1 · 2026-11-11
- PyMC Labs 发布营销基础模型 River:免逐数据集拟合的贝叶斯 MMM — twiecki · 2026-11-11
- NaiveAI 万 token/秒模型实测:最快但想得更久、话更多 — Latent Space · 2026-11-11
- TheZmi 观察:Fable 5.2 在多项鲁棒性测试上不升反降 — TheZvi · 2026-11-11
- Siri AI 端侧首层 Skimmer 拆解：纯正则免模型，最快比 LLM 快 20 倍 — SumitGup · 2026-11-11