Autor: Jakub Rusinowski · Ostatnia aktualizacja: 1 marca 2026
Flagowy model MoE Moonshot AI o bilionie parametrów, z 384 ekspertami i unikalną architekturą Agent Swarm. Napędza funkcję Composer 2 w IDE Cursor. Prowadzi w wielu benchmarkach programistycznych i rozumowania, w tym 93,33% na AIME 2026 i najlepszy wynik Pass@1 na SWE-rebench. Przy 32B aktywnych parametrach na token samodzielny hosting wymaga infrastruktury centrum danych.
| Licence | What it permits | Applies to |
|---|---|---|
Custom Open-Weight | Commercial use permitted Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms. | Kimi K2.5 1T |
| Kimi K2.5 1T | Min 605 GB VRAM · Q4_K_M · 256,000 ctx · |
Install Ollama then run: ollama run
Minimum VRAM: 605 GB. For best results use Q4_K_M quantization.
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.
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.
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.
Install Ollama, then run: ollama run . This downloads Kimi K2.5 and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.