My notes on Kimi K3, plus some thoughts on what we can still learn from the pelican benchmark even w...
TL;DR · AI 摘要
Moonshot AI发布2.8万亿参数的Kimi K3模型，但pelican基准测试已无法准确反映模型在实际工具调用等场景的真实表现。
核心要点
- Kimi K3参数量达2.8万亿，是当前参数规模最大的中文模型之一
- pelican基准测试与实际应用场景（如工具调用）存在显著脱节
- 模型评估应更关注实际对话中的代理工具调用能力而非单纯基准测试分数
结构提纲
按章节快速跳转。
介绍Moonshot AI发布的2.8万亿参数Kimi K3模型的基本信息
对比当前主流大模型参数量，强调Kimi K3的参数优势
分析pelican基准测试在评估实际工具调用能力时的不足
讨论当前模型评估体系与实际对话场景需求的差距
提出需要更关注实际对话中的代理工具调用等能力评估
思维导图
用一张图看清主题之间的关系。
查看大纲文本（无障碍 / 无 JS 友好）
- Kimi K3与pelican基准测试
- 模型参数规模
- 2.8万亿参数
- 当前最大中文模型
- 基准测试局限性
- pelican基准测试偏差
- 实际工具调用能力不足
- 评估方向调整
- 关注实际对话场景
- 强化代理工具调用评估
金句 / Highlights
值得收藏与分享的关键句。
Kimi K3拥有2.8万亿参数，是目前最强大的中文模型之一
pelican基准测试分数与模型在实际工具调用场景中的表现已出现显著偏差
当前模型评估体系需要更重视实际对话中的多轮代理工具调用能力
Simon Willison on X: "My notes on Kimi K3, plus some thoughts on what we can still learn from the pelican benchmark even while it becomes further detached from how good the models are at the things that matter (like agentic tool calling across longer conversations) https://t.co/F4AicvxAmq" / X
Simon Willison
@simonw
My notes on Kimi K3, plus some thoughts on what we can still learn from the pelican benchmark even while it becomes further detached from how good the models are at the things that matter (like agentic tool calling across longer conversations)
simonwillison.net
Kimi K3, and what we can still learn from the pelican benchmark
Chinese AI lab Moonshot AI announced Kimi K3 this morning, describing it as their “most capable model to date, with 2.8 trillion parameters”. It’s currently available via their website and …
8:24 PM · Jul 16, 2026
24.5K
Views
2
22
8
28
3
1
318
4
124
Read 22 replies