Event ·Multi-source

DeepSeek 发布 V4.1 Flash 模型，性能对标 GLM-5.3，KV 缓存技术大幅降低内存占用

DeepSeek 推出 V4.1 Flash 模型，其性能在测试中超过 V4 Pro，达到与 GLM-5.3、Kimi K3 相当的水平。该模型通过 CSA2 架构和 FP4 KV 缓存技术，将每 token 的缓存内存从 4KB 降低至 890 字节，内存占用减少 4 倍，使推理成本大幅下降。

Current read

DeepSeek 新出的 V4.1 Flash 模型，性能不错，和 GLM-5.3 差不多，关键是内存消耗特别低，适合做 agentic 任务，价格也便宜很多。

8 sources58° AI heatLast updated 2026/9/10 15:30:28

Evidence

8 sources

Primary sourceBlognone

2026/9/10 15:30:28View source

Related source 1掘金本周最热

2026/9/10 09:16:09View source

Related source 2arXiv cs.AI

2026/9/8 17:25:47View source

Related source 3OpenRouter

2026/9/9 14:57:26View source

Related source 4Cognition

2026/9/10 15:21:01View source

Related source 5The Rundown AI

2026/9/10 16:44:04View source

Related source 6IT之家

2026/9/11 10:03:41View source

Related source 7Fireworks AI

2026/9/8 18:21:50View 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