Event ·Multi-source

研究显示 AI 达到同等基准性能的成本每年下降约 13 倍

Epoch AI 的测算显示，AI 达到固定基准性能水平的价格每年下降约 13 倍，速度快于以往任何技术。MIT 研究在剔除硬件进步和竞争因素后，把算法层面的年度进步定在约 3 倍。价格下降不等于最强模型更便宜，推理模型每个任务消耗的算力更多，成本可能反而更高。实际选模型时，质量、速度和错误率与价格同样重要。

Current read

Epoch 算出同性能成本一年降 13 倍，MIT 拆出算法本身贡献约 3 倍，但推理模型反而更烧钱，选型逻辑讲得挺清楚。

5 sources58° AI heatLast updated 2026/9/24 16:00:18

Evidence

5 sources

Primary sourceDecoder

2026/9/24 16:00:18View source

Related source 1Epoch AI

2026/9/23 17:21:56View source

Related source 2AI Will

2026/9/26 01:38:07View source

Related source 3Thomas Wolf

2026/9/24 15:44:22View source

Related source 4IT之家

2026/9/26 10:18:15View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story