Kimi K2.5 — Moonshot AI 的本地 AI 模型
作者： Jakub Rusinowski · 最后更新：
Moonshot AI's 1-trillion parameter MoE flagship with 384 experts and a unique Agent Swarm architecture. Powers Cursor's Composer 2 IDE feature. Leads multiple coding and reasoning benchmarks including 93.33% on AIME 2026 and best Pass@1 on SWE-rebench. With 32B active parameters per token, requires datacenter infrastructure for self-hosting.
变体
Kimi K2.5 最小的变体在 Q4_K_M 下约需 605 GB 显存——量化权重加框架开销，不含 KV 缓存。
| 模型 | Q4 下显存 | 显存 | 上下文 | 运行 |
|---|---|---|---|---|
| Kimi K2.5 1T 1T | ~604.6 GB | 256,000 | ollama run kimi-k2-5 |
显存为 Q4_K_M 下的量化权重加开销，与 GPU 与显存检测器使用同一引擎计算。
如何在本地运行 Kimi K2.5
安装 Ollama，然后拉取标签。
ollama run kimi-k2-5在上方选择一个尺寸，查看它自己的显存、速度估算和安装命令。
许可证
Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms.
适用于: Kimi K2.5 1TKimi K2.5 — 常见问题
How much VRAM does Kimi K2.5 need?
Kimi K2.5 needs about 605 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: Kimi K2.5 1T (605 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run Kimi K2.5 on an RTX 4090 (24 GB)?
Kimi K2.5's smallest variant needs about 605 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
What quantization should I use for Kimi K2.5?
Q4_K_M is the best balance of quality and VRAM for Kimi K2.5 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run Kimi K2.5 with Ollama?
Kimi K2.5 has no local Ollama tag — the published tag is cloud-hosted, so running it sends your prompts to a hosted GPU rather than your own machine.