Primary source掘金本周最热
View sourceEvent ·Multi-source
Kimi K3.1 Pro 实测评测:100 万上下文与 3D 建模能力实测
作者用 Kimi Code CLI 界面版实测 Kimi K3.1 Pro,重点测了赛博上河图、天文机械表、超级玛丽复刻和甲维斯车库 3D 建模等案例。该模型为 2.8T MoE、100 万上下文、支持多模态,官方基准多数进入 top3,排位约在 Opus 5.5 之后、GPT-6 Astra 附近。实测中游戏复刻和 SVG 绘制明显优于 K3,也强于 DSV4.1_1017 和 MiniMax 3.1 等近期国产模型,但 3D 建模例子出现卡片错乱、轮子不滚动、车模不像 Mark III 等问题,与 GPT-6 Astra 仍有差距。文中还记录了指令理解失误(未按要求保存文件)等 Bad Case 和 token 消耗情况。
Current read
这篇不是通稿吹,作者是专门找 Bad Case 测的:K3.1 Pro 复刻超级玛丽、画机械表能打,3D 建模翻车现场也放出来了,想选模型的朋友看这个省得自己踩坑。
11 sources71° AI heatLast updated 2026/10/24 14:24:01
Evidence
11 sourcesRelated source 1kimmonismus
View sourceRelated source 2IT之家
View sourceRelated source 3rohanpaul_ai
View sourceRelated source 4TestingCatalog
View sourceRelated source 5Yuchen Jin
View sourceRelated source 6Simon Willison’s Weblog
View sourceRelated source 7Bindu Reddy
View sourceRelated source 8Artificial Analysis
View sourceRelated source 9Teortaxes
View sourceRelated source 10SiliconFlowAI
View sourceConflict 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.