---
id: 20261025-T0-01
title: "Kimi 3.1 Pro开源权重追平闭源模型，差距缩至1个月"
title_en: "Kimi 3.1 Pro Open Weights Narrow Closed-AI Gap to One Month"
url: https://ai.daily.yangsir.net/daily/20261025-T0-01
issue_date: 2026-10-25
publish_date: 2026-10-24T07:18:29.000Z
category: release
source_name: "Latent Space"
source_url: https://latent.space/p/ainews-kimi-31/
---

# Kimi 3.1 Pro开源权重追平闭源模型，差距缩至1个月

月之暗面发布Kimi 3.1 Pro，这是一个1040亿激活参数、2.8万亿总参数的开源权重模型，据称性能对标Fable-5.1级别，将开源与闭源模型的差距压缩到约1个月。同时推出的Pro Highspeed API速度提升2到3倍。所谓"Fable在家里"的说法来自付费推广方，实际能力仍需独立验证。对开发者而言，开源权重意味着可以自行部署和微调，高速API则适合对延迟敏感的在线应用。

## English Version

**Kimi 3.1 Pro Open Weights Narrow Closed-AI Gap to One Month**

Moonshot AI released Kimi 3.1 Pro, an open-weights model with 104B activated and 2.8T total parameters, claimed to match Fable-5.1-class performance and shrink the open-versus-closed gap to roughly one month. A Pro Highspeed API offering 2-3x faster inference launched alongside it. The "Fable at home" framing comes from paid promoters, so real capability still needs independent verification. For developers, open weights enable self-hosting and fine-tuning, while the high-speed API suits latency-sensitive applications.

---

**来源**：[Latent Space](https://latent.space/p/ainews-kimi-31/)

**详情页**：https://ai.daily.yangsir.net/daily/20261025-T0-01

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*